Intelligent Electric Vehicle Integration - Domain Interfaces and

General rights Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain • You may freely distribute the URL identifying the publication in the public portal

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

Downloaded from orbit.dtu.dk on: Apr 14, 2018

Intelligent Electric Vehicle Integration - Domain Interfaces and Supporting Informatics

Andersen, Peter Bach; Østergaard, Jacob; Poulsen, Bjarne; Gantenbein, Dieter

Publication date:2013

Document VersionPublisher's PDF, also known as Version of record

Link back to DTU Orbit

Citation (APA):Andersen, P. B., Østergaard, J., Poulsen, B., & Gantenbein, D. (2013). Intelligent Electric Vehicle Integration -Domain Interfaces and Supporting Informatics. Kgs. Lyngby: Technical University of Denmark (DTU).

http://orbit.dtu.dk/en/publications/intelligent-electric-vehicle-integration--domain-interfaces-and-supporting-informatics(dd87a91d-66b8-4b9c-b79b-a9ab5e7d66ae).html

Peter Bach Andersen

Intelligent Electric VehicleIntegration - Domain Interfaces

and Supporting Informatics

PhD Thesis, November 2012

mailto:[email protected]

Intelligent Electric Vehicle Integration - Domain Interfaces and SupportingInformatics

Author(s):Peter Bach Andersen

Supervisor(s):Prof. Jacob Østergaard, Technical University Of DenmarkAssoc. prof. Bjarne Poulsen, Technical University Of DenmarkResearch Scientist, Dieter Gantenbein, IBM Zurich Research Lab

Funding:EC power (1/3)Danish Agency for Science, Technology and Innovation (Forskningsstyrelsen) (1/3)Technical University of Denmark (1/3)

PhD school:Department of Electrical Engineering, Technical University of Denmark

Department of Electrical EngineeringCentre for Electric Power and Energy (CEE)Technical University of DenmarkElektrovej 325DK-2800 Kgs. LyngbyDenmark

www.cee.dtu.dkTel: (+45) 45 25 35 00Fax: (+45) 45 88 61 11E-mail: [email protected]

Release date: TBDClass: 1 (public)Edition: 1Comments: This report is a part of the requirements to achieve the PhD degree at

the Technical University of Denmark.Rights: c©DTU Electrical Engineering, 2012

ii

Abstract

This thesis seeks to apply the field of informatics to the intelligent integra-

tion of electric vehicles into the power system. The main goal is to release

the potential of electric vehicles in relation to a reliable, economically effi-

cient power system based on renewables.

To make intelligent EV integration a reality, it is prudent to understand

the domain in its entirety. In this thesis, this is reflected by a thorough

investigation of the stakeholders most relevant to the synergistic relationship

between electric vehicle and grid.

The first investigation addresses the power market. The market can give

system operators access to the flexibility of electric vehicles while at the

same time creating an immediate economic incentive for the EV owner. A

fleet operator is introduced to allow a fleet of electric vehicles to participate

in the markets. Examples are provided on the specific markets and services

in which the electric vehicle may be best suited to participate.

The next stakeholder investigated is the distribution system operator rep-

resenting the low voltage grid. The challenge is assessed by considering a

number of grid impacts studies. Next, a set of grid congestion mitigation

strategies are proposed with a special attention to the impact that conges-

tion would have on the operation of a fleet operator.

The third and most important stakeholder is the electric vehicle owner. The

emphasis is on the plug in patterns of a number of Danish electric vehicle

drivers. The objective is to understand how owner behavior will influence

charging flexibility. It is indicated how plug in behavior may be predicted

and how the resulting flexibility may be applied to achieve several different

goals.

After having investigated the aims, constraints and requirements for the

above stakeholders, the attention, in the second part of the thesis, is turned

to three vital topics within the field of informatics.

The first topic is the control architecture that determines the placement

and relationship between control systems used to control electric vehicle

charging. A centralised market-based architecture is chosen and the func-

tionalities needed by the control logic are defined.

The next informatics topic, communication, describes a set of protocols and

standards applicable for electric vehicle integration. The study investigates

the IEC 61850 standard and its ability to support smart charging.

Finally it is described how considerations to each of the stakeholders can

be included in the optimization done by the fleet operator. It is shown

how different markets can be considered and how stochastic optimization

can be used to model uncertainty in regards to plug in behavior and grid

congestion.

A large part of the above work have been done as contributions to the

EDISON project in which the Thesis Author has participated. During the

project the author has built a technical platform for testing several of the

technologies mentioned above, against a small fleet of electric vehicles. This

thesis is meant as an input for market players, system operators, fleet op-

erators, fellow researchers and anyone with an interest in the role of the

electric vehicle in the future power system.

Resume

Denne afhandling vil undersøge hvorledes informatik kan bruges til at un-

derstøtte en intelligent integration af elbilen i det danske elsystem. Formalet

er at frigøre elbilens potentiale i forhold til et palideligt og økonomisk opti-

meret elsystem baseret pa vedvarende energi. For at realisere en intelligent

integration er det nødvendigt, at undersøge og forsta omradet i dets helhed.

I denne afhandling er dette reflekteret i en bred undersøgelse af en række

hovedaktøre som vil have en særlig interesse i interaktionen mellem elbil

og elnet. Den første undersøgelse er rettet mod elmarket. Markedet kan

bade give systemperatører adgang til elbilens fleksibilitet samtidig med, at

elbilens ejer far en økonomisk kompensation. En fladeoperatør introduc-

eres som bindeled mellem elbiler og marked. Der gives specifikke eksempler

pa de undermarkeder og systemydelser som en elbil kan være bedst egnet

til at indga i. Den næste aktør er distributionsselskabet og det lavspænd-

ingsnet, som det repræsenterer. En række analyser af elbilers pavirkning

i denne del af elnettet er blevet undersøgt og en række strategier bliver

foreslaet for at imødega overbelastinger. Opmærksomheden er særligt rettet

mod hvorledes netbegrænsninger vil pavirke en fladeoperatør i dens ladeko-

ordinering. Den sidste og vigtigste aktør er elbilsejeren. Her undersøges el-

bilsejeres tilslutningsmønstre for at klarlægge elbilernes fleksibilitet i forhold

til intelligent opladning. Data fra en række danskeres brug af elbiler er brugt

i et studie og de første resultater fremlægges. Det indikeres ogsa hvorledes

tilslutningsmønstre kan forudsiges og den givne ladefleksibilitet bruges mod

forskellige mal. Den næste del af afhandlingen beskæftiger sig med dis-

cipliner indenfor informatik, og hvorledes de skal muliggøre integrationen

af elbiler i forhold til de hensyn, mal og begrænsninger, som er specifi-

ceret af ovenstaende aktører. Først behandles den kontrolarkitektur som

skal definere placeringen og forholdet mellem de systemer, som skal kon-

trollere en eilbils ladeadfærd. Der argumenters for brugen af en centralis-

eret markedsbaseret arkitektur og der gives eksempler pa de funktioner, som

den nødvendige styringslogik skal implementere. Dernæst beskives mulige

kommnikationsløsninger i form af en række protokoller og standarder. Dette

inkluderer en undersøgelse af IEC 61850 standarden og dens evne til at un-

derstøtte intelligent elbilsintegration. Det sidste emne inden for informatik

er den optimering af ladeadfærd, som en fladeoperatør vil udføre under

hensynstagen til de tidligere beskrevne aktører. Det beskrives hvorledes

usikkerhed i forhold til distributionsnettet og elbilsejeren kan inkluderes i en

stokastisk optimeringmodel, og hvorledes flere typer markeder kan benyttes

til at reducere prisen for opladning. En stor del af de ovenstaende emner er

lavet som bidrag til EDISON projektet i hvilket denne afhandlings forfatter

har deltaget. Under samme projekt har forfatteren udviklet en teknisk plat-

form, ved hjælp af ovenstaende teknologier, der bliver brugt til at demon-

strere intelligent opladning mod en række elbiler. Denne afhandling er ment

som input til markedsaktører, fladeoperatører, systemoperatører, forskere

og alle, som har en interesse i elbilens rolle i fremtidens elnet.

Acknowledgements

I would like to start by paying tribute to the parties which have provided fi-

nancial support to my PhD. The opportunity of pursuing a PhD and having

the financial backup to explore the field through international conferences

and external stays is a privilege not to be taken for granted.

I extend my heartfelt gratitude to my supervisors. To Bjarne, who origi-

nally introduced my to CEE and without whom I would not have written

this thesis. To Jacob, whose dedication to the field and his center have been

highly encouraging. To Dieter, whose passion for innovation is both conta-

gious and inspirational. To Chresten who, while not an official supervisor,

with his academic integrity and friendly nature has been a great support

throughout these three years.

To my colleagues and fellow researches from whom I have received a lot of

constructive input. Thank you, Salva, Ole, Andreas, Olle, Bernhard, Juan,

Shi, Kai, Junjie, Pierre and everyone at CEE.

To my close colleagues in our small EV smart grid integration group, Francesco

and Anders. I will always remember our small office where great ideas have

been born and plants have gone to die.

Finally I thank my family, my caring parents and wonderful wife. Thanks

for putting up with odd working hours and for politely listening to my

lengthy rants about my research.

To all of you: THANK YOU.

The LimerickPlug them in

Charge them up

From your own plug socket

Proudly put your foot down

As you drive around

In your electric rocket

Source: readingjuice.co.uk

ii

Contents

List of Figures vii

List of Tables ix

Glossary xi

1 Introduction 1

1.1 EV Integration - Exploring a New Paradigm . . . . . . . . . . . . . . . . 1

1.1.1 The Smart Grid . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.1.2 Distributed Energy Resources and Virtual Power Plants . . . . . 2

1.1.3 The Electric Vehicle and Utilization Concepts . . . . . . . . . . . 3

1.2 Quantifying the Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.2.1 Immediate Benefits to Environment and Health . . . . . . . . . . 5

1.2.2 Benefits Achieved trough Intelligent Integration . . . . . . . . . . 7

1.3 Projects and Coordinated Work Efforts within EV Research . . . . . . . 7

1.4 This PhD Study and its Contributions . . . . . . . . . . . . . . . . . . . 9

1.4.1 Thesis Objectives and Scope . . . . . . . . . . . . . . . . . . . . 9

1.4.1.1 Domain Interfaces . . . . . . . . . . . . . . . . . . . . . 10

1.4.1.2 Supporting Informatics . . . . . . . . . . . . . . . . . . 11

1.4.2 Thesis Documentation - Content and Structure . . . . . . . . . . 12

1.4.2.1 Publication list . . . . . . . . . . . . . . . . . . . . . . . 13

1.4.2.2 Thesis Structure . . . . . . . . . . . . . . . . . . . . . . 16

2 Market Integration 19

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.2 Related Work Within Area . . . . . . . . . . . . . . . . . . . . . . . . . 20

iii

CONTENTS

2.3 Thesis Work and Contributions . . . . . . . . . . . . . . . . . . . . . . . 21

2.3.1 Market Integration in EDISON . . . . . . . . . . . . . . . . . . . 22

2.3.2 Comparative Study between Ancillary Service Markets . . . . . . 23

2.3.3 Providing Ancillary Services in Denmark . . . . . . . . . . . . . 26

2.4 Sub-conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

2.4.1 Contributions and Results by the Thesis . . . . . . . . . . . . . . 28

2.4.2 Discussion and Recommendations . . . . . . . . . . . . . . . . . 29

3 Grid Integration 31

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.2 Related Work within Area . . . . . . . . . . . . . . . . . . . . . . . . . . 33


3.4 Sub-conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37



4 EV Owner 41

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41



4.3.1 Simple Prediction based on Driving Data . . . . . . . . . . . . . 45

4.3.2 Early Investigation of Real Driving Data . . . . . . . . . . . . . . 47

4.3.3 Plug in Patterns: Seasonal Changes . . . . . . . . . . . . . . . . 49

4.3.4 Plug in Patterns: Familiarization . . . . . . . . . . . . . . . . . . 50

4.3.5 Plug in Variance Case and Relation to Smart Charging . . . . . 52

4.4 Sub-conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53



5 Control Architecture 57

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57



5.3.1 System Architecture of EDISON . . . . . . . . . . . . . . . . . . 61

iv

CONTENTS

5.3.2 High-level Functions . . . . . . . . . . . . . . . . . . . . . . . . . 62

5.4 Sub-conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64



6 Communication 67

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67



6.3.1 Research on the IEC 61850 . . . . . . . . . . . . . . . . . . . . . 70

6.3.1.1 Using REST Services . . . . . . . . . . . . . . . . . . . 71

6.3.1.2 Data Model . . . . . . . . . . . . . . . . . . . . . . . . . 72

6.3.1.3 Test of IEC 61850/REST Implementation . . . . . . . . 73

6.4 Sub-conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75



7 Optimization 79

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79



7.3.1 Cost Minimization . . . . . . . . . . . . . . . . . . . . . . . . . . 82

7.3.2 Introducing Markets . . . . . . . . . . . . . . . . . . . . . . . . . 83

7.3.3 Introducing Uncertainty - EV Owner Behavior . . . . . . . . . . 85

7.3.4 Introducing Uncertainty - Grid Constraints . . . . . . . . . . . . 87

7.4 Sub-conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89



8 Conclusions 91

8.1 Domain Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

8.2 Supporting Informatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

8.3 Intelligent EV Integration - Closing Remarks . . . . . . . . . . . . . . . 96

v

CONTENTS

9 Core publications 97

A Coordination Strategies for Distribution Grid Congestion Management

in a Multi-Actor, Multi-Objective Setting . . . . . . . . . . . . . . . . . 98

B A comparison of electric vehicle integration projects . . . . . . . . . . . 107

C Smartly charging the electric vehicle fleet . . . . . . . . . . . . . . . . . 115

D Implementation of an Electric Vehicle Test Bed Controlled by a Virtual

Power Plant for Contributing to Regulating Power Reserves . . . . . . . 144

E Prediction and optimization methods for electric vehicle charging sched-

ules in the EDISON project . . . . . . . . . . . . . . . . . . . . . . . . . 152

F Facilitating a generic communication interface to distributed energy re-

sources - Mapping IEC 61850 to RESTful Services . . . . . . . . . . . . 160

References 167

vi

List of Figures

1.1 Controlled charging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.2 Active power service . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.3 Energy backup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.4 CO2 emission comparison . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.5 Domain overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.6 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.1 Market integration comparison . . . . . . . . . . . . . . . . . . . . . . . 20

2.2 Market integration in EDISON . . . . . . . . . . . . . . . . . . . . . . . 22

2.3 Smart charging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.4 Table-To-Grid setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2.5 EV setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.1 Map of base-case operations . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.2 Advance capacity allocation . . . . . . . . . . . . . . . . . . . . . . . . . 37

4.1 Expected night availability as function of variance . . . . . . . . . . . . 45

4.2 Implementation of predictions . . . . . . . . . . . . . . . . . . . . . . . . 46

4.3 Full sample of vehicles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.4 Plug in periods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

4.5 Winter vs. summer month . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.6 Initial vs. mature use . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4.7 SOC distribution for case EV . . . . . . . . . . . . . . . . . . . . . . . . 52

4.8 Plug in/out cumulative distributions and input for value functions . . . 54

5.1 Centralized and distributed architecture . . . . . . . . . . . . . . . . . . 58

vii

LIST OF FIGURES

5.2 Function-based design for a GVPP . . . . . . . . . . . . . . . . . . . . . 60

5.3 Comparison of architectures . . . . . . . . . . . . . . . . . . . . . . . . . 60

5.4 Architecture in EDISON . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

5.5 High-level centralized fleet operator functionalities . . . . . . . . . . . . 63

6.1 Project comparison of communication technologies . . . . . . . . . . . . 69

6.2 The 61850 standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

6.3 Mapping to URL path . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

6.4 Logical nodes used in implementation . . . . . . . . . . . . . . . . . . . 73

6.5 Communication setup in EDISON . . . . . . . . . . . . . . . . . . . . . 74

6.6 Test of IEC 61850 communication . . . . . . . . . . . . . . . . . . . . . 74

7.1 Optimization approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

7.2 Markets and stages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

8.1 Chapter topic relationships . . . . . . . . . . . . . . . . . . . . . . . . . 96

viii

List of Tables

2.1 A/S market comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.1 Plug in period averages . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

4.2 Plug in period seasonal averages . . . . . . . . . . . . . . . . . . . . . . 50

4.3 Plug in period familiarization averages . . . . . . . . . . . . . . . . . . . 51

7.1 Symbol overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

ix

GLOSSARY

x

Glossary

A/S Ancillary Service

CVPP Commercial Virtual Power Plant

DER Distributed Energy Resource

DSO Distribution System Operator

ENTSO European Network of Transmission

System Operators for Electricity

EV Electric Vehicle

EVSE Electric Vehicle Supply Equipment

FDR Frequency controlled Disturbance

Reserve

FNR Frequency controlled Normal opera-

tion Reserve

GAMS General Algebraic Modeling System

GVPP Generic Virtual Power Plant

ICE Internal Combustion Engine

ICT Information and communication

technology

IEC International Electrotechnical Com-

mission

OEM Original Equipment Manufacturer

PLC Power Line Communication

REST REpresentational State Transfer

RTO Regional Transmission Organization

SLA Service Level Agreement

SOA Service Oriented Architecture

SOAP Simple Object Access Protocol

SOC State-Of-Charge

TLS Transport Layer Security

TSO Transmission System Operator

TVPP Technical Virtual Power Plant

V2G Vehicle-To-Grid

VPP Virtual Power Plant

xi

GLOSSARY

xii

1

Introduction

1.1 EV Integration - Exploring a New Paradigm

Since Scottish inventor Robert Anderson built the first electric vehicle (EV) in 1832, the

world has seen several attempts at electrifying the transportation sector. Successful EV

introduction represents a challenge of daunting proportions but with an extraordinary

potential for society. A challenge which can only be solved through a large set of

disciplines spanning such fields as anthropology, politics, business, and technology.

In the following, the term “EV Integration” will be used to cover the technological

challenges and describe the solutions that engineers and technicians working within

electric components, power system design, informatics, and mathematics have to com-

bine in a number of cross-disciplinary solutions. EV Integration research should, for

instance, find solutions for:

• Infrastructure

• Roaming

• New EV services

• Smart grid integration

The goal of the latter area would be to minimize the adverse effects of introducing

EVs as a new load in the power system and, instead, promote them as an active asset.

This would ultimately reduce society’s cost, monetary as well as environmental, for

the electrification of transportation. To achieve the above the EV itself would have

1

1. INTRODUCTION

to be equipped with capabilities so that it through information and communication

technology (ICT) can receive stimuli from its surrounding environment and through

logic adjust its charging or discharging behavior.

The main EV integration challenge addressed in this thesis is that of intelligent EV

integration into the power system as part of the smart grid concept.

1.1.1 The Smart Grid

The “smart grid” is said to be a trend of huge proportions in the Western world, a

significant revolution in the world of power and energy, a major paradigm shift in

markets, and one of the biggest engineering challenges of the century. It is, however,

also complex, early in its development and may have to be carried out in iterations as

part of an evolution.

The standard organisation “IEEE” gives the following definition: “The ”smart grid”

has come to describe a next-generation electrical power system typified by the increased

use of communications and information technology in the generation, delivery and con-

sumption of electrical energy.” (1)

Academia has put forward countless suggestions as to what this increased use of

ICT may be used for. Practical applications pursued at the moment include intelligent

metering, home automation, distributed production from Photovoltaics (PV), micro

combined heat and power units (CHP), and the control of loads such as air conditioning

(A/C), electric heating and EV charging.

The latter type of units, decentralized production and controllable loads, is often

grouped under the umbrella of a common term: distributed energy resources (DER).

DER is geographically distributed units with a resource potential. DER integration

constitutes a large part of smart grid research.

1.1.2 Distributed Energy Resources and Virtual Power Plants

As more DERs connect to the various levels of the power system, a new challenge for

the system operators - DSOs and TSOs - emerges. An interesting feature of DERs,

however, is that a sufficient degree of controllability and intelligence can elevate them

from being a challenge to the grid to, instead, being an asset. This raises the question

as to how and to what end DERs should be controlled. While the purposes typically

explored in academia relates to cost reduction, environmental and/or socio-economic,

2

1.1 EV Integration - Exploring a New Paradigm

it remains a question what purpose will bring about the greatest value. One answer to

the question as to how to coordinate the behavior of DERs, one viable solution is the

use of a Virtual Power plant (VPP)

A VPP describes an aggregated system in which many DERs are partly or fully

controlled by a single coordinating entity. For the individual DER, the VPP can hide

the complexity of market interactions and help meet market requirements that the

individual DER would be too small to accommodate. For the power system operators

and market players, the VPP aggregator could hide the complexity of interfacing with

a host of small individual units. The name “Virtual power plant” explicitly refers to a

VPP aggregator’s ability to mimic a traditional large power plant in behavior. VPPs

have been researched as part of the European FENIX project (2) and in academic

publications such as (3, 4).

When applied to EV integration the term “fleet operator” is often used for the

aggregating entity that controls the DERs as part of the VPP concept. This use of the

VPP concept can be found in the EDISON project and is described in a publication

co-authored by the Thesis Author (5). In this thesis both the term “fleet operator” and

“aggregator” will be used. A “fleet operator” can be said to describe the role taken by

a “real” actor operating in the current market and business space while “aggregator”

is closer to the more generic academic definition of the same entity. Currently, busi-

nesses in Denmark engaged in providing charging-related services to EV owners uses

terms such as “Electric Mobility Operator”to describe themselves. Chapter 5, “Control

architecture” will go into more detail with the structure of a VPP.

1.1.3 The Electric Vehicle and Utilization Concepts

While the EV is similar to other DERs according to the above definition, some of its

properties make it of particular interest. The main defining properties of an EV that

elevates it potential as a DER is that it is:

1. Expected to be grid-connected and available most of the time with a high degree

of flexibility

2. A high-power, quick-response flexible load with an attached storage and possibly

with bi-directional power flow capabilities

3

1. INTRODUCTION

For the first property the following can be observed. As other types of household

attached DERs, such as CHP, Wind, and roof-installed solar, the unit is acquired and

installed to meet a purpose separate from its potential functions in a smart grid context.

For the EV this primary function is transportation which is also the main limiting

factor in the degree to which the EV can be used for smart grid related matters. It has,

however, been established that a typical vehicle is parked for as much as 23 hours a

day in a publication by Kempton (6), and that these periods, especially at night, follow

certain patterns that are predictable to an extent which allow EV charging to pursue

other purposes than reaching a adequate energy level for the next drive. Kempton

therefore describes the EV as “a vastly underutilized resource”.

The second property owes itself to the fact that the DER in question is built for

transportation; i.e., it uses a drive train system capable of transferring large quantities

of power to and from the battery on short notice. Generally speaking, it does not matter

whether energy is transferred back and forth between the battery and motor due to

acceleration and regenerative braking or between battery and grid as part of smart

grid integration. It is also likely that batteries will become better at accommodating

shallow mid-State-Of-Charge cycles. Having established the EV as a DER with a high

grid potential, the remaining question is this: what should this potential be used for.

In the remainder of this thesis the term “intelligent EV utilization concept” is meant

to cover all services and goals that the manipulation of the direction, rate and timing

of the power and energy exchanged between vehicle and power grid can accommodate.

The following introduces three EV capabilities that can each support a range of

utilization concepts for various levels of the power system.

Controlled charging Charging is delayed or advanced in time based on, e.g.,

energy cost or renewable contents. A specific utilization concept could be following

a dynamic energy tariff. Controlled charging is often described by the term “Smart

charging”.

Figure 1.1: Controlled charging - Illustrative example of controlled charging behavior

4

1.2 Quantifying the Benefits

Active power services Short-duration charging and, possibly, discharging opera-

tions. This capability can be used towards ancillary services - either the ones present

in the current power market or new types of ancillary services aimed at the local dis-

tribution net.

Figure 1.2: Active power service - Illustrative example of active power service behavior

Energy backup Using a portion of the EV battery capacity to store energy which

is to be delivered back to the surrounding power system at a later point in time. This

capability can be used towards the utilization concept of storing wind energy at hours

with low demand.

Figure 1.3: Energy backup - Illustrative example of energy backup behavior

The above types of behavior can be employed in either system-wide or local grid

utilization concepts. Examples of the former is energy backup for transmission grid

attached renewables and energy market prices. Local grid concepts could be voltage

control, congestion mitigation, and co-existence with other DERs. This thesis will

touch upon the above utilization concepts throughout this thesis.

1.2 Quantifying the Benefits

The benefits that the introduction of electric vehicles can accommodate can be di-

vided up between immediate environment and health benefits and the additional socio-

economic benefits that can be obtained through smart grid integration and EV intelli-

gence.

1.2.1 Immediate Benefits to Environment and Health

On a local level, EV introduction will reduce particle emission and street noise. Both of

these pollutants are proven to have adverse effects on the public health. Noise pollution

5

1. INTRODUCTION

have been investigated by WHO in (7) and particle pollution by Aarhus University in

(8). Besides from the intangible human consequences, the result is an increase in the

cost of public health care and a decrease in workforce productivity.

On a national level the introduction of EVs can reduce oil dependency and green

house gas emission, respectively, thereby reducing human-made contributions to global

warming. The latter is true not only when considering the pollution-free electric motor,

but also when considering how the energy, on which it runs, was generated. The term

“Well-To-Wheel” C02-emission encompasses the sum of produced CO2 by all parts of

the energy production chain.

Based on a Well-To-Wheel calculation, a typical EV would produce less than half

the CO2 than a traditional ICE car in the same size-class as seen in figure 1.4.

Figure 1.4: CO2 emission comparison - The EV emission is based on C02 productionof energy generation in western Denmark (0,449 g CO2/Wh), data source: Dansk ElbilKomite (9)

Electrifying the entire Danish transport sector would reduce the national C02 emis-

sion considerably. This means that the integration of EVs is crucial to meeting the

milestones/targets of 2020, 2030 and 2050 for green house gas emission issued by the

Danish government.

6

1.3 Projects and Coordinated Work Efforts within EV Research

1.2.2 Benefits Achieved trough Intelligent Integration

A unique feature of the EV is that it, after the point of acquisition, can become “cleaner”

over time. This is true if the renewable component of a nation’s power generation

increases over time and especially if the intelligence of the EV promotes and accelerates

an increased penetration of renewable production.

The Danish TSO, EnergiNet.dk, have released a report on efficient use of wind

energy (10) stating the following:

”The analysis shows that heat pumps and EVs have a big potential in creating a

new and large flexible energy demand which mean that a larger portion of wind power

can be used inside Denmark. Thereby Co2-emissions will be reduced considerably”

Another benefit related to intelligent EV integration is the socio-economic costs of

implementing and operating a next-generation power system.

The general cost of the future power system has been investigated in another En-

erginet.dk report (11) which estimates a DKK 6.1 billion net-earning for Denmark if

smart grids are implemented. The report stresses that the services provided by”green

and flexible vehicles” are essential for realizing this potential. Energinet.dk report (10)

even more explicitly mentions EV integration. From the report summary:

”It is furthermore of vital importance that solutions are developed that supports an

intelligent interaction between electric vehicles, heat pumps and the power system..”

”..If this intelligence is not developed in the communication between power system

and the new flexible demand, the socio-economic benefit of electric vehicle and heat

pump integration will be reduced by approximately DKK 1.7 billion per year.”

There is therefore good reason to promote the introduction of EVs in society and

pursue an intelligent integration into the power system.

1.3 Projects and Coordinated Work Efforts within EV

Research

Over the last couple of years, EV integration has received a lot of attention from both

academia and industry. A lot of research groups have conducted and published research

which is both relevant and valuable to EV integration. This section will mention some

of the research groups and projects which have made significant contributions to the

field

7

1. INTRODUCTION

The EDISON project (12), “Electric vehicles in a Distributed and Integrated market

using Sustainable energy and Open Networks”, was a research project whose goal it was

to develop optimal solutions for EV integration, including network issues, market solu-

tions, and optimal interaction between different energy technologies. The project has

thoroughly investigated a long list of issues relating to EV introduction and integration.

The e-Mobility Berlin project have made significant contributions to the field (13).

The project was initiated by Daimler AG (Mercedes-Benz) and the utility RWE. Among

the participants are also battery, Electric Vehicle Supply Equipment (EVSE) and other

auto-mobile original equipment manufacturers (OEM). The project introduced a fleet

of 100 EVs supplied by Daimler and 500 EVSEs which were delivered and powered by

RWE in the streets of Berlin for a large field test. The project was primarily aimed

at developing and testing standardized solutions for electric vehicles and contributed

towards infrastructure standardization in Europe.

The Vattenfall Wind-to-vehicle (W2V) project (14) was a cooperation initiative

between BMW and the energy company Wattenfall. The project deserves mention

for being among the first to demonstrate the ability of a group of EVs to participate

in a system-wide utilization concept - charging based on wind production. For the

demonstration, a complete ICT system was set up to coordinate a number of “mini-e”

vehicles. The same group has moved on to do grid impact studies and to demonstrate

that the coordination of EVs may also benefit the distribution grid.

The MERGE project (15) was launched as an “evaluation of the impacts that

EV will have on the EU electric power systems regarding planning, operation and

market functioning.”. Among the notable contributions of this project was its grid

impact studies that in great detail, and for different topologies and EV integration

levels showed the impact of EV introduction on European grids. The projects also

defined a stage-based roadmap for the realisation of EV smart charging schemes, short

and long-term.

A research group lead by Professor Willett Kempton, University of Delaware, has

made great contributions to EV research (16). Among the contributions are the demon-

stration on market integration via Vehicle-To-Grid technology and the economic as-

sessment of regulation power provision. Technically, the group has contributed with

solutions in the form of control systems and power electronics for EV monitoring and

8

1.4 This PhD Study and its Contributions

control as well as communication solutions and the application of agent theory in the

coordination of EV charging.

IBM researchers Carl Binding and Olle Sundstroem have worked on optimization

methods to support smart charging. The researchers have considered a wide range of

relevant parameters such as grid congestion issues, EV owner behavior, market trading,

and battery lifetime issues. To this end they have used a equally long array of meth-

ods such as optimization (linear, quadratic etc.), grid modelling (load flow, network

model etc.), battery models, and statistical techniques. The publications from these

researchers constitute a great contribution to the field in general and have been an

inspiration for parts of this thesis.

The International Electrotechnical Commission (IEC) have made standards relevant

to EV integration(17), some of the most significant have been mentioned in several of

the core publications of this thesis. The standards put forward by IEC can greatly

influence the shape and direction of the field in Europe. The above is also true for the

European Committee for Electro technical Standardization (CENELEC) (18).

The above groups and projects has in several cases produced academic results which

are useful as input to this thesis. These inputs will be described in the “related work”

sections in each of the following chapters.


This section outlines the exact goals of this study and the contributions made to address

them. The objectives pursued by the Thesis Author are described in the “Thesis

objectives and scope” section and the “Thesis documentation - content and structure”

section describes the written contributions that documents how the PhD author have

sought to meet the objectives. This is done by listing all the papers and book chapters

that have been published and then map a chosen subset - the core publications - to the

structure of this main thesis report.

1.4.1 Thesis Objectives and Scope

Solving any type of engineering challenge can be split into two basic steps: Firstly, a

thorough investigation of the domain in which the challenge exist is done, and, sec-

9

1. INTRODUCTION

ondly, the best possible solutions are recommended. This basic, but very fundamental,

approach allows for the definition of the two main objectives of this thesis:

1. To investigate the main stakeholders within intelligent EV integration and the

objectives, considerations and constraints they each represent.

2. To identify and find the best contemporary informatics technologies to solve the

challenge of intelligent EV integration.

By addressing the above objectives the main contribution of the thesis becomes an

investigation of the domain in which the EVs will be integrated and the identification

of appropriate informatics to carry out such an integration.

In the end, the thesis is meant to consolidate the EVs role in the ultimate, over-

arching mission of smart grid research: the development of a reliable, sustainable and

economic power system.

This thesis divides the work done into two parts. The first part describes the domain

investigation and the second deals with the investigated informatics solutions. Each

part is described in the following sections.

1.4.1.1 Domain Interfaces

As more and more EVs connect to the power system, the foundation for relationships

and dependencies to a new set of stakeholders is established.

Entities like fleet and system operators will have business models, objectives and

constraints that may influence the charging behavior of an EV in many different ways.

These new relationships should be implemented in a way that is financially beneficial,

convenient and acceptable to the EV owner.

The domain considered in this thesis is grid-centric with a focus on three levels.

First there is the consumer-level representing the individual grid connection point.

This level is included since understanding the behavioral patterns, value drivers and

requirements of prosumers, in this case EV owners, is a clear necessity for EV integra-

tion. Next, the distribution level is included due to the very real operational challenges

and opportunities that EV introduction may entail in this part of the system. Finally,

on the system-wide level, the market is included to represent system-wide energy and

10


power services since the market may drive the economic incentives behind the various

intelligent EV utilization concepts.

The domain along with the levels described above is shown in figure 1.5

Figure 1.5: Domain overview - Main stakeholders and objectives within intelligent EVintegration

The stakeholders considered by this thesis, as part of the domain interfaces inves-

tigation becomes:

• Market player (Market integration)

• Distribution system operator (Grid integration)

• EV owner

Each of the above stakeholders is covered by a chapter in this thesis.

1.4.1.2 Supporting Informatics

Informatics, as part of computer science, covers the structure, algorithms, behavior,

and interactions of natural and artificial systems that store, process, access, and com-

municate information. The thesis will apply part of the academic field of informatics

to the specific application of EV integration.

11

1. INTRODUCTION

The topics selected cover the major aspects of the information value chain: archi-

tectural system relationships between entities collecting and exchanging data; the ways

in which the data is formatted and sent as part of the communication process between

said entities; and, finally, the ultimate goal of predicting future events and achieve value

through optimization.

The supporting informatics topics of this thesis becomes:

• System architecture

• Communication

• Optimization

Each of the above topics are covered by a chapter in this thesis.

This thesis and its publications have attempted to illuminate how the domain study

of EV owner, grid and market integration have impacted recommendations within

choice of technologies. The conclusions chapter will illustrate this relationship.

1.4.2 Thesis Documentation - Content and Structure

This main thesis document, made up by the following chapters and sections, is publication-

based, i.e., this “enveloping document” lists and references a set of publication made

throughout the PhD study and relies on them as a direct part of the thesis work doc-

umentation.

Each chapter have primarily been defined to cover a meaningful portion of the

overall PhD thesis objectives, but will also be closely linked to one or more publications.

Since most of the documentation is left to the publications, the main objective of

the following chapters will be to summarize findings, relate the work to that done by

others, add possible corrections, supply more context and perspective, and finally to

relate each topic to the field of intelligent EV integration in its entirety.

Two exceptions are chapter four “EV owner” and chapter seven “Optimization”

where new material has been included which has not been documented elsewhere. The

following sections will describe the contributions made by the author in terms of pub-

lications and project contributions, list a number of core publications with special

relevance to the thesis and finally map them to the thesis structure.

12


1.4.2.1 Publication list

A subset of the publications by the Thesis Author has been selected as core publications.

A core publication is a written contribution with special relevance to the objectives

defined by the thesis. The letter next to the publication will be used as a identifier

throughout the document.

APeter Bach Andersen, Junjie Hu, and Kai Heussen.Coordination strategies for distribution grid congestion manage-ment in a Multi-Actor, Multi-Objective Setting. In Proc. 3rd IEEEPES Innovative Smart Grid Technologies (ISGT) Europe, Berlin, 2012.

BPeter Bach Andersen, Rodrigo Garcia-Valle, and Willett Kempton.A Comparison of Electric Vehicle Integration Projects. In Proc. 3rdIEEE PES Innovative Smart Grid Technologies (ISGT) Europe, Berlin, 2012.

C

Peter Bach Andersen, Einar Bragi Hauksson, Anders Bro Pedersen, DieterGantenbein, Bernhard Jansen, Claus Amtrup Andersen, and Jacob Dall.Smartly charging the electric vehicle fleet. In book “Smart Grid Ap-plications, Communications, and Security”, Wiley-VCH (ISBN: 978-1-1180-0439-5), pages: 381-408, 2012.

D

Francesco Marra, Dario Sacchetti, Anders Bro Pedersen, Peter Bach Andersen,Chresten Træholt, and Esben Larsen.Implementation of an Electric Vehicle Test Bed Controlled by aVirtual Power Plant for Contributing to Regulating Power Reserves.In Proc. IEEE Power and Energy Society General Meeting, San Diego, 2012.

E

Andreas Aabrandt, Peter Bach Andersen, Anders Bro Pedersen, Shi You,Bjarne Poulsen, Niamh O’Connell, and Jacob Østergaard.Prediction and optimization methods for electric vehicle chargingschedules in the EDISON project. In Proc. IEEE PES Innovative SmartGrid Technologies (ISGT), (ISBN: 9781457721588), Washington, D.C., 2012.

F

Anders Bro Pedersen, Einar Bragi Hauksson, Peter Bach Andersen, BjarnePoulsen, Chresten Træholt, and Dieter Gantenbein.Facilitating a generic communication interface to distributed energyresources - Mapping IEC 61850 to RESTful Services. In Proc. FirstIEEE International Conference on Smart Grid Communications (SmartGrid-Comm), (ISBN: 978-1-4244-6510-1), Gaithersburg , 2010.

13

1. INTRODUCTION

The core publications are all publications to which the Thesis Author have con-

tributed with a significant part of the work. An exception is the core publication D by

Marra, which have been included since parts of the demonstrations described consti-

tute proof-of-concept of work done by the author. All core publications are appended

to this document. The Thesis Author have also made contributions to a number of

other publications not directly included in or referenced by this thesis. They are not be

included since the Thesis Authors role is smaller than in the core publication and/or

the topic they address falls outside the scope of this thesis document. The following is

the full list of publications to which the author have made contributions, excluding the

core publications. (Begins on the following page...)

14


Anders Bro Pedersen, Peter Bach Andersen, Joachim Skov Johansen, David Rua,Jos Ruela, and Joo A. Peas Lopes.ICT solutions to support EV deployment. In Book “Electric Vehicle Integra-tion Into Modern Power Networks”, Springer, 2013.

Shi You, Junjie Hu, Anders Bro Pedersen, Peter Bach Andersen, Claus NygaardRasmussen, and Seung-Tae Cha.Numerical Comparison of Optimal Charging Schemes for Electric Vehi-cles. In Proc. Proceedings of IEEE PES General Meeting, San Diego, 2012.

Qiuwei Wu, Arne Hejde Nielsen, Jacob Østergaard, Seung-Tae Cha, FrancescoMarra, and Peter Bach Andersen.Modeling of Electric Vehicles (EVs) for EV Grid Integration Study. InProc. 2nd European Conference SmartGrids & E-Mobility, (ISBN: 978-3-941785-14-4), Brussels, 2010.

Carl Binding, Dieter Gantenbein, Bernhard Jansen, Olle Sundstroem, Peter BachAndersen, Chresten Træholt, Francesco Marra, and Bjarne Poulsen.Electric Vehicle Fleet Integration in the Danish EDISON Project. InProc. IEEE Power & Energy Society General Meeting, (ISBN: 978-1-4244-6549-1),Minneapolis, 2010.

Peter Bach Andersen, Bjarne Poulsen, Chresten Træholt, and Jacob Østergaard.Using Service Oriented Architecture in a Generic Virtual Power Plant. InProc. Sixth International Conference on Information Technology: New Generations(ITNG ’09), (ISBN: 978-1-4244-3770-2), page: 184, Las Vegas, 2009.

Morten Decker, Bjarne Poulsen, Peter Bach Andersen, Cresten Træholt, and JacobØstergaard.Theoretical evaluation of a generic virtual power plant framework usingservice oriented principles. In Proc. 6th Mediterranean Conference and Ex-hibition on Power Generation, Transmission, Distribution and Energy Conversion,Thessaloniki, 2008.

Peter Bach Andersen, Bjarne Poulsen, Morten Decker, Cresten Træholt, and JacobØstergaard.Evaluation of a Generic Virtual Power Plant Framework Using ServiceOriented Architecture. In Proc. 2nd. IEEE International Conference on Powerand Energy (PECCon 08), (ISBN: 978-1-4244-2404-7), pages: 1212-1217, Johor Ba-haru, 2008.

15

1. INTRODUCTION

Furthermore, the Thesis Author has made contributions to the following projects:

EDISON The Thesis Author has been in charge of DTUs participation in EDISON

Work Package 3 : “Distributed integration technology development” and has made

written contributions to the technical reports of the project.

• (Internal report) Existing Asset Analysis.

• (Internal report) Distributed reference architecture.

• (Public report) D3.1 Distributed integration technology development.

The Thesis Author have developed large parts of the back-end system - The EDISON

Virtual Power Plant (EVPP) - used for demonstrations throughout the project.

SEESGEN ICT The Thesis Author has contributed to the SEESGEN-ICT Work

Package 4: “ICT for Demand Side Integration” Here the Thesis Author made written

contributions to the reports:

• (Public report) D4-2: Report on ICT requirements, offers and needs for Demand

Side Integration.

• (Public report) D4-3: ICT for Demand Side Integration - Report on Technical

and non-Technical barriers and solutions.

Having listed the previous contributions by the Thesis Author, the next section

outlines this thesis document and its structure.

1.4.2.2 Thesis Structure

The structure of this document is made to reflect the two main objectives of the thesis.

This means that the thesis consist of two parts: ”Domain interfaces” and ”Support-

ing informatics”. The ”Domain interfaces” part consists of chapters reflecting the key

stakeholders identified and thus consists of chapters dealing with the EV owner, dis-

tribution grid and market. In the same way, the ”Supporting informatics” part holds

three chapters covering control architecture, communications and optimization.

Each chapter will start with a list of the core publication that contribute to the

specific topic.

16


The thesis document structure, along with the publication to chapter mapping, is

illustrated in figure: 1.6.

Figure 1.6: Thesis structure - Thesis structure with core publications mapped tochapters

The thesis document will conclude with a “Conclusions” chapter that will present

and discuss the results, make recommendations, and finally indicate areas in need of

further investigation.

17

1. INTRODUCTION

18

2

Market Integration

This chapter deals with the market as part of the “Domain interfaces” investigations

of this thesis. The following publications constitute the core contributions done by the

Thesis Author to this area:

B A comparison of electric vehicle integration projects.

D Implementation of an Electric Vehicle Test Bed Controlled by a Virtual Power Plant

for Contributing to Regulating Power Reserves.

E Prediction and optimization methods for electric vehicle charging schedules in the

EDISON project.

The publications will be referred to by their letter (B,D,E) in the following. Part

of the material in this chapter, presented in the “Comparative study between ancillary

service markets” section, is part of a special project report “Ancillary services and

EV aggregator operation” done by the Thesis Author under supervision of Professor

Willett Kempton at the University of Delaware. The report has not been published,

but is available upon request.

2.1 Introduction

The existing power and energy markets may represent the easiest short-term way of

integrating fleets of Electric Vehicles (EV) into the power system. Adjusting charg-

ing according to energy prices or participation in markets product can both benefit the

19

2. MARKET INTEGRATION

power system and represent an immediate economic incentive for fleet operator and EV

owner. Furthermore, the study of current market conditions provides a good oppor-

tunity for shaping the recommendations and requirements for future markets to come.

For the above reasons this thesis, and projects such as EDISON and the University of

Delaware’s Vehicle-To-Grid (V2G) programme, uses existing markets for the investi-

gations on EV utilization concepts. There are, however, shortcomings in the current

markets that need to be addressed as part of DER market integration research: Firstly,

requirements in terms of minimum bid sizes may represent a barrier for market entry.

This issue was one of the main reasons the Virtual Power Plant (VPP) concept was

originally proposed. The need for such an entity may decrease as the market evolves,

but is, at least for now, a necessity for not only meeting capacity limits but also to

manage the complexity of market involvement on behalf on individual EV owners. The

use of VPP aggregation vs. distributed control is covered in more detail in chapter five

“Control Architecture”. Secondly, since power markets were not built for DERs such

as EVs, the flexibility and services of such units may not be appropriately rewarded if

the mechanisms and products of the markets are left unchanged. A lot of attention is

therefore given to designing the future markets that will better facilitate the integration

of DERs. This is the main purpose of projects such as EcoGrid.EU (19). Finally, the

distribution grid may be subject to new kinds of markets. This may help ensure a fair,

safe and efficient utilization of the LV network. This is touched upon in chapter three

“Grid integration” in this thesis.

2.2 Related Work Within Area

As part of the core publication, B, three projects was investigated in terms of their

approach to market integration. See figure 2.1

Figure 2.1: Market integration comparison - Comparison of the market integrationapproach of three EV projects

20

2.3 Thesis Work and Contributions

As can be seen in the figure, University Of Delaware (UD) V2G is the only of the

three projects that actively investigated the aspect of interfacing with the regulation

market. The work by UD V2G within advanced EV utilization deserves special men-

tion in this section due to their focus on this more advanced market approach and

their practical demonstrations of active participation in such a market. Market par-

ticipation was achieved by forming a VPP-like constellation and by partnering with a

large battery-based regulation provider. The description of these demonstrations and

the associated economics incentives is found in (20, 21). While UD V2G is unique in

terms of the results they have achieved, numerous other researchers have investigated

integration with either spot markets, ancillary service (A/S) markets or a combination

of the two. In the paper (22), Rotering uses price forecasts and dynamic programming

in a demonstration of both spot market based charging cost minimization and A/S

provision. In a case based on markets in California, it is shown how daily charging cost

can be reduced from $0,43 to $0,2 . The daily profit of regulation service participation

is estimated to be $1,71. In the article (23), Kristoffersen describes a fleet operator

charging according to spot marked prices while considering typical driving patterns.

The article suggests using either linear or quadratic programming to plan charging,

depending on whether the fleet operator can influence market prices i.e. being a price

taker vs having market power. In a report by Heesche (24), the author estimates the

charging costs based on different charging strategies and mix between the spot and

regulation market. Charging costs can, for some night charging scenarios, be cut in

half when using smart charging instead of “dumb” charging. The estimated savings,

however, is less than EUR 10 per month and are unlikely to be large enough to in-

centivize smart charging. Especially when considering that the market energy price is

typically only a small fraction of what the EV owner pays per kWh. In Denmark only

15-25 percent of the energy bill is the energy price, the rest is taxes and fees.

Letting the tax component of the energy price increase and decrease as a function of

the time of use, can strengthen the economic incentive of smart charging in the future.


Core publications part of the PhD study have described how the charging of a fleet of

EVs can be based on hourly spot market prices or regulation signals. The following sec-

21


tion covers work done as part of the EDISON project, a comparison between American

and European A/S products and, finally, a few demonstrations on A/S provision.

2.3.1 Market Integration in EDISON

As previously shown, EDISON choose to focus on existing markets and in this case on

the Nordic Spot day-ahead energy market. The first step taken was an investigation

of how a Fleet operator would be introduced in the setup of the spot market. The

following figure 2.2 illustrates the approach used.

Figure 2.2: Market integration in EDISON - The introduction of a fleet operator inthe EDISON project

The figure shows some of the existing market roles (Retailer, production- or load

balance responsible) to which an actor, such as the fleet operator, would be associated.

In a real market scenario the fleet operator would bid on energy based on historic

prices and the expected demand of its fleet. For the demonstrations done in EDISON

however, the bidding process was neglected and it was shown how charging cost could

be minimized if coordinated directly with known spot day-ahead energy prices. Rather

than spot market integration it can be argued, this reflect a scenario where a dynamic

energy tariff is used for charging.

Figure 2.3 shows a piece of demonstration software made in EDISON and illustrates

how charging would be moved to the cheapest hours during night.

22


Figure 2.3: Smart charging - The charging schedule (bottom graph) places charging inhours where prices are low (middle graph).

For charging planning, linear optimization was used. More information is available

in the core publication E and in chapter seven “Optimization”. Since energy prices

are generally fixed for EV owners there are no immediate economic incentive for smart

charging as done in the optimization. If energy is purchased through a retailer trough

the spot market it could, however, mean a lowering of charging prices. Using the spot

prices directly in the cost estimation does not give a accurate impression of the real

savings an end user would see. Rather this would depend on the bidding strategy of

the fleet operator/retailer.

2.3.2 Comparative Study between Ancillary Service Markets

A type of markets that could offer higher savings, or even a profit, for the EV owners

and fleet operators, are the A/S markets.

As mentioned previously in this chapter, UD V2G has already proven that there

is a potential for connecting a fleet of EVs with the regulation market in the PJM

area. To better understand both the requirements, potential barriers and type and

shape of economic compensation that different markets represents, an investigation has

been made. The goal was to compare the conditions under which UD V2G successfully

demonstrated market integration with the conditions in the Danish A/S and regulation

markets as part of European Network of Transmission System Operators for Electricity

(ENTSO-E). The summery of this investigation can be seen in table 2.1.

23


A/S Market Area Controlmethod

Contracting Minimumbid size(MW)

Responsivenessrequirement(To full ca-pacity)

Capacitycom-pensa-tion

Average capacity pay-ment (MW-Hr)

Energy payment(MW)

Frequency reg-ulation

PJM(USA)

Controlsignal

Day ahead,One-hourblocks

0,5 5 Min Energy - 12,5 EUR A Based onRMCP and in some casesLOC

SynchronizedReserve

PJM(USA)

Controlsignal

Day ahead,One-hourblocks

0,5 10 Min Energy - 7,3 EUR B Tier 1: Basedon LMP plus a pr MWpremium Tier 2: Basedon SRMCP and in somecases LOC

Manual regula-tion reserves

DK west+ DKeast

Manual Day ahead,One-hourblocks

10 15 Min Capacityand en-ergy

1,97 EURC up west

0,08 EURC down west

0,49 EURC up east 0,86

EURC down east

N/A

Primary Re-serve

DK west Frequencysensing

Day ahead,Six-hourblocks

0,3 30 Sec Capacity 50,4 EUR D up 16,85

EUR D down

Secondaryreserve

DK west Controlsignal

Purchasedon a monthlybasis

N/A 15 Min Capacityand en-ergy

N/A Based on spot- and in-traday regulation marketprices.

Frequencycontrolled nor-mal operationreserve

DK east Frequencysensing

Purchasedon a monthlybasis

0,3 2,5 Min Capacity 78,36 EUR E up 58,87

EUR E down

Frequency con-trolled distur-bance reserve

DK east Frequencysensing

Day ahead,Six-hourblocks

0,3 30 Sec Capacity 31,06 EUR F up

Table 2.1: A/S Market comparison - Comparison between ENTSO-E and PJM marketsA 2010, annual report, Regulation Market clearing price, including opportunity cost source:www.monitoringanalytics.com B 2010, annual report, Tier 2, Mid-Atlantic Subzone of the RFC SynchronizedReserve Market source: www.monitoringanalytics.com C 2010, average based on market data extract, source:www.energinet.dk D 2010, average based on market data extract, source: www.energinet.dk E 2010, averagebased on market data extract, source: www.energinet.dk F 2010, average based on market data extract, source:www.energinet.dk

By comparing the A/S markets listed in the table, the following observations can

be made. A/S contracting works in pretty much the same way across the ENTSO-

E and PJM A/S markets. The Transmission System Operator (TSO) or Regional

Transmission Organization (RTO)(Role for PJM) will choose from a set of bids for

up and down regulation reserves, the providers will be informed about accepted bids

and the final compensation will be settled. The length of the time interval for which

the A/S is contracted, and the time-of-contracting, however, varies from case to case.

These differences should mean little to a fleet operator.

Most of the ENTSO-E markets described here are frequency controlled. An ex-

ception is the secondary reserve of DK west which is somewhat comparable with the

regulation market of PJM in regard to activation mechanism i.e. a positive or negative

power set-point sent from the TSO/RTO to the provider. If a provider using an EV

fleet wish to participate in the frequency controlled A/S markets of DK west and DK

east it will need to use frequency sensing equipment as part of the control system.

24


This equipment will have to satisfy certain requirements on accuracy and measurement

resolution.

Power and responsiveness requirements differ somewhat from market to market. For

Frequency Controlled Disturbance Reserve (FDR) and primary reserve, the minimum

bid size is 0,3 MW while the other markets investigated operate between 1 MW to

10 MW. This makes FDR and primary reserve suitable for participation by smaller

EV fleets in the early stages of EV integration. The responsiveness requirements of

the markets are all easily satisfied by an EV fleet. A demo V2G vehicle as described

in (21) was able to respond on a sub-second timescale with its full capacity where as

the fastest Danish A/S markets listed in this report requires full capacity within 30

Seconds.

The payment in the A/S markets generally increase as a function of how much

power can be supplied in how short a time. Common for most ENTSO-E A/S markets

described here is that they only reward capacity and not energy. The manual regula-

tion reserve does not seem suitable for EV fleets since it offers a poor compensation

for capacity and requires substantial amount of energy to be available for continuous

supply. It is argued in (21) that it is currently less economically feasible to supply

larger amounts of energy back to the grid due to roundtrip losses and a relatively low

compensation pr. kWh in most markets.

The remaining four A/S markets in Denmark have much higher compensation.

This is especially true for Primary Reserve and Frequency controlled Normal operation

Reserve (FNR). FDR offers a lower average capacity compensation than FNR even

though the former has higher requirements to responsiveness. This can be attributed

to the fact that FDR is invoked less often. FDR can, in this aspect, be compared to

synchronized reserves in PJM and could offer the same trade-off between compensation

and EV battery lifetime preservation.

As previously mentioned, bid size requirements can be meet by the use of an aggre-

gated group of vehicles. The concept of using multiply resources to meet a demand is

mentioned in the “combined delivery” sections of the “Ancillary services to be delivered

in Denmark - Tender conditions, valid from 1 January 2011” manual of Energinet.dk.

Here it is stated that a delivery can be made up from several production or consump-

tion units with different properties which collectively can provide the required response

within the required response time.

25


For several of the markets, however, a delivery is not allowed from a combination

of producing and consuming units. It will therefore be relevant to establish if an EV

delivering power back to the grid (V2G) will be considered a “producer”. Will, for

instance, a fleet operator be allowed to meet an up regulation request by making some

EVs reduce load and having others deliver power back.

2.3.3 Providing Ancillary Services in Denmark

Having investigated A/S market participation by EVs, the Thesis Author participated

in two demonstrations where two kinds of physical units were made to react on market

signals.

Based on the market investigation, the secondary reserve market of DK was chosen

since the regulation sum signal, provided by Danish TSO Energinet.DK, is both a

meaningful and easily implementable input to the test. The secondary reserve signal

is a MW value sent from the TSO to a group of providers and is positive or negative

depending on the need of either up or down regulation. The sum signal is all the

individual signals added together and is used since individual “TSO to provider” signals

are kept confidential.

The first test was performed towards a battery setup as described in the core publi-

cation D. The test demonstrated how a set of lithium batteries through a controllable

charger and inverter, could be made to charge or discharge according to the real-time

balancing need of the TSO. The result can be seen in figure 2.4.

The upper graph in the figure shows the charging schedule sent to the battery

setup compared to the target value (sum signal) sent by the TSO. For this test Marra

D constructed and prepared the physical setup (nicknamed the “Table-To-Grid” setup).

The lower graph in the figure shows measurements from the battery management system

(BMS) attached to the batteries.

The Thesis Author decided to repeat the experiment, using the full system architec-

ture and communication solutions described later in this thesis, towards a real EV (the

Toyota Scion eBox). The exact same signal was sent to the EV as was used towards

the battery-setup. The result can be seen in figure 2.5.

It can be seen from the figure that the EV precisely follows the aggregated regulation

signal. The regulation signal in question represents a period with many fluctuations

with a close to equal amount of up and down regulation periods. This has a positive

26


Figure 2.4: Table-To-Grid setup - Using a set of lithium batteries to respond to a LFCsum signal sent by a TSO. source: Marra, core publication D

Figure 2.5: EV setup - Using an EV to respond to a LFC sum signal sent by a TSO

27


impact on the State-Of-Charge (SOC) which stays close to 50 percent throughout the

whole period. A stable energy level makes planning easier when having to meet a

certain energy target plus reduces wear on the battery due to deep cycles. For periods

with less variation, e.g., longer periods dominated by either up or down regulation, an

aggregator will have to depend on having sufficient vehicles in its portfolio as to even

out the influence on individual EV SOC levels.

The experiments prove that A/S via EVs is technically possible - proving the busi-

ness case and understanding the impact on batteries still is in need of research.

2.4 Sub-conclusion

The following sections concludes this chapter by presenting and discussing results and

by giving recommendations on future work.

2.4.1 Contributions and Results by the Thesis

The first findings and contributions, as part of the EDISON project, were that:

• EVs can be connected to the NordPool spot day-ahead energy market through a

VPP-based fleet operator if adhering to the appropriate market roles and respon-

sibilities.

• Backend software have been developed and have been used to charge EVs (real

and simulated) according to energy prices extracted from the energy market.

The next step taken was looking at A/S markets to understand how the market

used in UD V2G compares with the ones available in Denmark, the results were that:

• Although variations exists between the markets, most constraints and require-

ments can be overcome by the use of a fleet operator and the aggregation of

EVs.

• The FNR market of DK east is found to offer the highest average capacity com-

pensation but would require frequency sensing to be implemented.

• A/S based on power capacity rather than energy, is a good match for the capa-

bilities of an fast-response, high-power DER such as an EV.

28

2.4 Sub-conclusion

• The secondary reserve market of DK west is a close match to frequency regulation

in the PJM area and is a good starting point for a field test.

• The EnergiNet.dk rules on “combined delivery” will have to accommodate V2G

enabled EVs.

Based on the findings from the market analysis the following was demonstrated:

• One of the Danish markets were chosen and it was demonstrated that two units

- a Table-To-Grid setup and an eBox EV - can follow the sum signal sent out by

the TSO as would be the case when participating in said market.

• The experiments carried out by the Thesis Author have proven that an EV

can be made to react within a few seconds (full communication stack latency

+ charger/inverter ramp-time) which would easily satisfy all markets considered.

2.4.2 Discussion and Recommendations

Market integration concepts for EVs will vary based on local market conditions and

are, for the centralized architectures, tightly linked to the business cases chosen by the

fleet operator. Smart charging, as covered by EDISON and e-mobility, can be seen as a

first approach to market integration by allowing the EV to respond to variable energy

prices on the energy market. The V2G equipment tested by UD V2G can be seen as

the second step where the EV increases its potential as a resource to the grid and more

market integration concepts becomes possible. New market integration concepts might

occur over the coming years to better facilitate DERs as part of the smart grid concept.

It is recommended to further examine the FNR and Primary reserve markets since

they, at the time of the investigation, are found to offer the best economical com-

pensation. This would require new functionalities to be implemented in the technical

platform used by the fleet operator in order to facilitate frequency sensing as required

in these markets.

It is important to note that the economic attractiveness of a specific market can

change over time. If, for instance, several EVs enter a specific market they might, in

time, saturate the need of certain services. Also, nightly energy prices may increase as

more EVs delay their charging to the same hours. As long, however, as most European

29


countries keep their ambitious energy targets, it is likely that the flexibility of EVs will

be both needed and duly rewarded on the markets.

This chapter have described the use of the energy market for a charging cost re-

duction and the A/S market for a potential profit. An advanced scheme could be to

combine both of the above objectives, i.e., planning EV charging based on both energy

prices and balancing needs. This is left for future research.

30

3

Grid Integration

This chapter deals with the distribution grid as part of the “Domain interfaces” in-

vestigations of this thesis. The following publications constitute the core contributions

done by the Thesis Author to this area:

A Coordination Strategies for Distribution Grid Congestion Management in a Multi-

Actor, Multi-Objective Setting.


EDISON project.

The publications will be referred to by their letter (A,E) in the following.

3.1 Introduction

The local distribution grid, to which the Electric Vehicle (EV) will connect, has a large

part to play within intelligent EV integration and may be one of the most complex

issues that EV integration research will have to address. The core challenge is that

the EV represent a larger load than what is normally attached to residential feeder

lines and can therefore violate the Distribution System Operators (DSO) operational

constraints such as:

• Voltage limits. Keeping the voltage at any point in the grid within a certain

range (power quality obligation towards customer).

31

3. GRID INTEGRATION

• Thermal limits of cables and transformers. The energy transferred through

equipment and conductors compared to the current rating of such components

(equipment lifetime consideration).

• MVar limits. Bands for reactive power exchange at the connection between

the distribution grid and the transmission level (obligation towards transmission

system operator).

In this chapter the focus is on congestion issues, i.e., a situation in which the demand

for active power transfer exceeds the transfer capability of the grid.

Where charging according to market prices may be the first obvious approach when

talking about system-wide concepts (See chapter two “Market integration”), moving

the EV load away from the household peak load is an obvious first approach when

turning the attention to the distribution grid and congestion issues. It has, however,

been established that the combined load of EVs in themselves can cause problems for

the grid even if charging when base-load is typically low. Add to this that many of the

system-wide concepts considered will tend to cluster charging in time e.g. all reacting

to the same low charging cost or reacting to a regulation need at a higher level of the

power system. This behavior is by the IBM partners in the EDISON project termed

”The EV lemming effect”. (25)

Making the EV charging “compatible” with the local grid can either be implemented

as a hard requirement imposed on fleet operators, or as a new type of LV grid ancillary

service which a fleet operator will be compensated for supplying (but still may be forced

to provide).

When researching EV grid integration, the three following steps can be followed:

1. Conducting grid impact studies to assess the problem.

2. Suggesting strategies to mitigate grid issues.

3. Validating strategies using simulations and in-field tests.

The Thesis Author have sought to understand issues relating to the first of the above

points, grid impact studies, by going through a number of publications investigating this

topic. The emphasis of this thesis has been on the second point, suggesting strategies

to mitigate grid issues (in the form of congestion), and this primarily from the point

32

3.2 Related Work within Area

of view of a fleet operator. The Thesis Author’s main motivation is to understand the

kinds of impact that grid constraints may have on the implementation and operation

of an aggregator and the informatics on which it relies.

In the end the objective is to find out to what degree system-wide services can be

harmonized with local grid constraints.


The EDISON project has made contributions to the area of grid impact studies which

is documented as part of the projects technical reports (26). Here, a part of the

Bornholm grid has been simulated and the result of different EV penetration levels was

investigated in terms of cable and transformer loading as well as voltage drops for part

of the 400 V, 10 kV and 60 kV grid. The conclusion was that the 400 V grid was most

sensitive with voltage drops as the biggest problem. All grids could accommodate up

to 20 percent EV integration if the charging was limited to 1 phase/ 16 Amp / 230 V

charging (3,6 kW). The charging strategy (“dumb” charging vs. smart charging) was

shown to have a major impact on the EV integration level supported. An important

comment relating to the fleet operator made by the report is ”The fleet operator has

the ability to limit its consumption, however, with the current market setup, there are

no incentives to do so.”.

Another project, MERGE, have performed comprehensive investigations on the

topic (27) and concludes that for LV networks ”large scale EV integration affects quality

of service and may cause significant technical problems”. MERGE bases the study on

several EV penetration scenarios, grid topologies and charging methods. MERGE sim-

ulations indicate that congestion problems increase when considering ”Dumb” charging

and rural grid topologies.

The effect grid topology has on simulation results is also visible when comparing

publications done by Shao (28) and Marre (29) respectively.

Shao demonstrates that a strong urban distribution grid supports a 100 percent EV

integration considering both voltage limits, harmonics and thermal loading of cables and

transformer. In contrast Marra demonstrates that the transformer of a rural/residential

topology can be overloaded at relatively low EV integration levels.

33

3. GRID INTEGRATION

It can be concluded from the above that grid impact is a real issue and should be

handled accordingly. This is especially true when considering the provision of system-

wide market services done by V2G UD (20) (using an EV fleet for regulation services),

Here EV charging will both be clustered in time and to be done a high power levels,

i.e., 120 V, single phase with up to 80 Amps (9,6 kW) thus adding to a potential grid

issue.

The next step for academia is to suggest congestion mitigation strategies. Also in

this area some good contributions are available.

In (30) Shao describes a demand response strategy for load shaping considering the

constraints of the local transformer. The strategy also includes considerations to user

preferences, the priority between the individual loads in a household and the privacy

of data. It is shown how household base load plus controllable load (AC, Water heater,

EV) will exceed transformer ratings in certain hours. A home area network control

flowchart illustrates how the controllable loads are scheduled depending on their priority

and convenience preference. By employing the strategy it is shown how the combined

load of the households is kept within the transformers capacity rating.

As part of the EDISON project, dynamic grid tariffs were investigated. In this

scheme, a price is sent that can reflect system-wide energy prices while including a

cost-component representing the state of the grid to which the individual recipient is

connected. This means that even if energy prices are low, a local price may be high due

to local congestion. The use of dynamic grid tariffs have been described by Connell in

(31).

Another notable contribution to congestion mitigation strategies combined with

smart charging has been done by Binding and Sundstroem in (32). The authors uses

discrete and continuous state variables to simulate a fleet of electric vehicles and the

distribution grid. The paper introduces a Charging Service Provider (CSP) which goal

it is to meet EV owner energy demand, minimize charging energy cost and at the same

time avoid grid bottlenecks. The solution may depend on a number of iterations be-

tween a DSO and the CSP before the fleet charging plan of the CSP is “pre-approved”.

The DSO uses a load flow or flow Network method to validate the impact of a certain

fleet charging plan sent to it by a CSP. The solution represent an advanced approach

to handling congestion issues.

34


The latter two of the above publications seek to combine grid congestion mitigation

with the minimization of charging costs. This type of combined objectives is an ad-

vanced and very relevant area of EV integration research and is also the area to which

this thesis wish to contribute.


The focus of this section is on how grid constraints may influence the operation of the

fleet operator.

Ideally a grid congestion mitigation strategy should be implemented that respects

and supports the operational goals of all stakeholder involved in EV integration. The

key stakeholders identified in core publication A is the DSO, the fleet operator and EV

owner. The approach used is to list the key operational tasks that each stakeholder

would need to undertake based on their value drivers. These tasks are then divided

into stages based on in what phase they are performed. Se figure 3.1.

Figure 3.1: Map of base-case operations - Mapping of stakeholder key operations tostages.

The stages used are inspired by functional levels found within industrial automa-

tion. “Planning” is the stage in which commitments and contracts are made. In

the “Scheduling” stage, available resources are known and they are coordinated and

prepared for the operation stage. The stage “Operation” is about pure execution of

35

3. GRID INTEGRATION

schedules and the handling of real-time events. Finally the “Settlement” stage is about

the financial aftermath.

The purpose of the map is to illustrate the task of each participant before a conges-

tion mitigation strategy is introduced. One special operation ”Charging interruption”

found in the “Operation” stage is suggested since it could represent a vital last resort

function for the DSO to stop the charging of an EV if all other grid congestion miti-

gation strategies fail. The suitability of a strategy could then, among other things, be

measured on the amount of times it would be necessary to invoke such a function.

Core publication A then describe the concept behind three individual strategies

which are:

• Distribution grid capacity market

• Advance capacity allocation

• Dynamic grid tariff

For each approach, a new map is drawn where operations required specifically for

the strategy in question are presented. The dynamic grid tariff strategy is similar to

what is proposed by Connell (31), i.e., generate a time and grid-location dependent

price for grid usage based on expected nodal consumption levels.

The distributed grid capacity market strategy introduces a ”market operator” which

defines prices based on demand and supply. Aggregated schedules for demand and

prices are sent back and forth between market operator and fleet operators in an iter-

ative process.

The final strategy, advance capacity allocation, the one described and proposed by

the Thesis Author, is illustrated in figure 3.2.

The idea behind this approach is that the DSO would calculate available capacity for

each feeder-line based on transformer and cable ratings and the expected conventional

load curves. The households attached to a certain feeder-line would then be given a

certain share of available capacity (kW), which could be allocated to the fleet operator

representing them. To avoid inefficient utilization of available grid capacity due to

unused capacity shares, a second step is added to the strategy where fleet operators

can trade their allocated capacity in an over-the-counter manner.

36

3.4 Sub-conclusion

Figure 3.2: Advance capacity allocation - Operational tasks and interactions for the“advance capacity allocation” strategy. Key operation “boxes” for this strategy are bolded.

Possible advantages and drawbacks of each strategy is discussed in A in terms of

complexity, value and risk.

A benefit for the advance capacity allocation strategy is that it would be relatively

easy to implement. The strategy, however, relies on efficient and reliable trading be-

tween fleet operators in order for the capacity to be used efficiently. The strategy does

not deal with real-time operational requirements such as harmonics and voltage limits

which needs to be handled separately.

A very early attempt to incorporate grid constraints in EV charging is found in core

publication E where the charging power can be limited based on a time based constraint.

This time-based constraint could be based on the capacity allocation strategy.

3.4 Sub-conclusion




A number of grid impact studies have indicated that congestion issues may occur at

EV penetration levels as low as 20-40 percent depending on charging strategy, charging

power and the topology and level (e.g. 400 V vs 10 kW) of the grid in question.

37

3. GRID INTEGRATION

The Thesis Author have contributed to the area with an investigation on grid con-

gestion mitigation strategies in which it is concluded that:

• Operation maps can help understand the impact of coordination strategies on the

operations and interactions of stakeholders.

• Grid considerations may have to co-exist with other objectives in the charging

management of EVs (such as system-wide services).

• Coordination and information exchange in earlier stages can reduce complexity

and benefit both the fleet operator and DSO.

• There may be a trade-off between ease of strategy implementation and optimality.

A compromise may be necessary for the first real-life implementations.

• A suitable strategy for coordination between DSOs and fleet operators will im-

prove each stakeholders ability to reach its objectives considerably.

If a strategy such as advance capacity allocation is used, the available per-feeder

capacity may be known by the fleet operator when planning EV charging in correspon-

dence with market contracting. These capacity limits can be included as constraints in

an optimization method.


Full penetration of EVs will undoubtedly mean extending and enforcing the grid in

some locations. If EV charging is coordinated appropriately, however, such extensions

may not need to accommodate a worst-case scenario where all EVs are charging at the

same time.

An important question is what objective will have the greatest monetary value:

using EVs for system wide services or to charge according to local grid conditions.

Short term, with the existing distribution grids, the DSOs operational limits will

trump any objective that a fleet operator may have in the power market. For the

planning of future distribution grids, however, the many possible system-wide uses of

the EV must be thoroughly considered.

Should distribution grids be subject to heavy extensions so that they can facili-

tate a lot of EVs charging simultaneously in response to e.g. transmission level wind

38

3.4 Sub-conclusion

availability!? or should expansions, and the associated cost, be kept at a minimum by

requiring all EVs to charge according to available grid capacity.

It is reasonable to assume that the DSO will require an real-time emergency in-

tervention mechanism that curtails EV charging if some of the operational limits are

violated. It would represent a big challenge to EV introduction in general if such events

are allowed to occur frequently.

39

3. GRID INTEGRATION

40

4

EV Owner

This chapter deals with the Electric Vehicle (EV) owner as part of the “Domain in-

terfaces” investigations of this thesis. The following publications constitute the core

contributions done by the Thesis Author to this area:

A Coordination strategies for distribution grid congestion management in a Multi-

Actor, Multi-Objective Setting.


EDISON project.

The publications will be referred to by their letter (A,E) in the following.

This chapter also present some material not yet published in the ”Early investigation

of real driving data” section. While the work in this area is not yet complete, it is

included here to supply some early examples on how real EV usage data may be used

by a fleet operator in its operation. The data can be used as input to the prediction

and optimization method employed by a fleet operator.

4.1 Introduction

The EV owner is easily the most important stakeholder in EV integration. Intelligent

EV integration depends both on the EV owner accepting her or his role as an active

prosumer and that the behavioral patterns of said owner support flexibility in charging.

41

4. EV OWNER

Regarding the first point, owner acceptance, it is important to understand the

motivations of EV owners. Early research (33) point to three main parameters that

influence people’s willingness to invest in an EV:

1. The EV’s ability to meet transportation needs - especially in regards to range.

2. The total cost of ownership compared to a traditional car.

3. The availability of sufficient EV charging infrastructure.

The intelligent EV utilization concepts investigated as part of this thesis, can po-

tentially influence the second of the above factors - total cost.

For any software developed to achieve controlled charging, it is reasonable to assume

that owners would like to have interfaces (mobile apps, web pages etc.) through which

they can stay informed and empowered. Also, the owner is more than likely to favor

convenience and ease of use.

Such convenience can be achieved trough:

1. Transparent frame conditions for the use of controlled charging captured in a well-

formed Service Level Agreement between owner and fleet operator. (Example:

always charge immediately to 50 percent energy for emergency trips)

2. That the fleet operator can precisely predict the needs and flexibility of the EV

owner and approach the owner with accurate and relevant suggestions for con-

trolled charging. (Example: The fleet operator can suggest that charging can be

delayed until morning hours if the owner never leaves home before noon)

The second of the above points, assessing the precise flexibility of the EV owner, is

also important in order to maximize the EVs value as a resource. The flexibility relates

to EV owner plug in patterns which are defined as the periods in time in which the

EV is connected to the power system and thus capable of uni or bi-directional energy

transfer. While the EV owner may plug in the vehicle at many different locations and

for different need scenarios, only one type of plug in period is of interest for smart grid

integration:

• A period that is long enough, compared to the energy need, so that they grant

flexibility for delaying charging in time.

42


• A period that is predictable in nature and recurs according to a certain pattern.

In general, two categories of commuter charging suit the two above criteria: work-

place charging and night charging. In contrast, most fully-public charging options (such

as DC fast charging) are used for range extension, i.e., used sporadically to gain extra

kilometers of range. The latter category therefore does not meet either criterion.

A general expression of the flexibility can be formulated as:

Flexibility = f(duration, energyNeed, predictability) (4.1)

The work presented in the thesis investigates plug in periods in terms of the above

parameters and it is shown how such periods can be predicted and what role they may

play for intelligent EV utilization concepts and the fleet operator.


Most driving pattern studies conducted have relied on data from traditional internal

combustion engine (ICE) cars for their investigations. This is due to the fact that any

large quantity of driving data collected from EVs are, as of yet, difficult to come by.

ICE vehicle data is a reasonable starting point if EVs are expected to fulfil the same

role as the ICE vehicles investigated, e.g., urban commuting.

Using ICE vehicle data can, for instance, be useful in studies investigating an EVs

ability to meet driving needs. Comparing a typical EVs range of 100-180 km with the

average Danish driving distance by commuters of 29,48 km (Danish National Transport

Survey) it seems that most daily driving patterns can be satisfied. It is, however, also

necessary to consider the variance e.g. how often people drive much longer than the

29,48 km.

In (34) Pearre conducts a statistical study based on 484 ICE vehicles to asses a

vehicle owners range needs compared to what would be supported by an EV. Pierre

approaches the problem by investigating how many times, over one year, the range

of an EV would not suffice. It is shown that nine percent of the sampled vehicles

never exceeds a daily driving need of 160 km, meaning that this group would be viable

candidates for adopting EVs. Pierre then shows that allowing six “range violations”

per year would increase this number to 32 percent. This means that if roughly one third

43

4. EV OWNER

of the sample group would find an alternative solution, e.g., rent an ICE vehicle, for

the up to six days with the longest driving distance, an EV could satisfy the remaining

driving need for this entire group.

Having investigated an EVs ability to meet the primary function, driving, the at-

tention can be turned towards the EVs potential for intelligent EV utilization concepts.

A study by Wu (35) address daily driving patterns as done by Pierre, but also

investigates EV availability for controlled charging. Here ”availability” is interpreted

as all the time periods in which the vehicle is not driving. The sample presented by

Wu shows a very high degree of availability - especially at night and during noon. This

availability, if representing times in which an EV is both parked and plugged in, is an

early first measure of the degree to which it can participate in concepts such as smart

charging.

The investigation of such “availability periods” is taken further in (36). Here, the

authors uses a semi-Markov model to predict the next place of arrival and stay duration

of a vehicle. The arrival locations are shown to be predictable with 84 percent accuracy

using the proposed method. As with the above studies, ICE vehicle data is used and

are based on GPS measurements.

The authors clusters geographically close GPS measurements for when the vehicle

is parked to create a number of locations. The importance of each location, in terms

of usefulness for controlled charging, is then estimated using a point system. Points

are given depending on the number of stopovers at the location, the average stopover

duration and the spread of this duration. The method is very useful when going beyond

night charging for EVs and identifying other periods that are sufficiently long and

predicable. For private commuting vehicles this would, however, most likely be limited

to the home and workplace location. Hidden Markov models have previously been

applied to trip prediction by Simmons in (37) which also addresses the specific purpose

of a trip.

There are, however, investigations for which you need ”real” EV data. For instance

when it comes to understanding how often the EV owner will perform the action of

plugging in the vehicle. Some of the largest organised EV trails yielding such data,

are the German Mini E Berlin project and the Danish Test-An-EV programme. An

investigation of EV owner acceptance and behavior based on Mini E Berlin is described

by Cocrona in (38)

44


The research on owner patterns based on real EV trails and the impact on EV fleet

charging optimization is still in need of investigation and is therefore pursued by the

Thesis Author.


As stated above, this thesis focuses on the use patterns of the EV owner as they relate

to smart charging. It is especially important that the fleet operator understands and

adheres to driving demands. See core publication A for a short description of expected

EV owner requirements and value drivers.

4.3.1 Simple Prediction based on Driving Data

The first study was based on ICE vehicle patterns as part of the AKTA driving set

(Centre for Traffic and Transport (CTT)). The AKTA data was based on a total of 360

vehicles, which were followed from 14 to 100 days between 2001 and 2003 and is biased

towards driving seen in the greater Copenhagen area.

It was, in core publication E, suggested to use a simple exponential smoothing

model for predicting the duration of night stopovers ( see figure 4.1).

Figure 4.1: Expected night availability as function of variance - Illustration ofhow the period used by a fleet operator for controlled charging may depend on the averagevalue and standard deviation of plug ins and outs.

In the driving data, stopover periods were periods in time where the car would be

parked. The authors filtered the data to only include nightly stopover periods and some

outliers were removed. These periods could then be considered suitable candidates for

smart charging and where used as input to a binary optimization (described in chapter

seven: ”Optimization”). The following exponential smoothing function was used:

45

4. EV OWNER

Xn+1 = (1− δ)Xn + δSn (4.2)

In the above, X1 = S1 and S ∈ Rn is historic stopover data for a given vehicle.

δ ∈ [0; 1] is a fixed parameter determining the weight that new observations are given

compared to historic data. A low δ-value will result in new observations having a higher

impact on the prediction.

The paper went further by investigating the relationship between the variance and

mean of start and stop times and the value of the objective function and to test the

robustness of the solution. A suggestion of the authors was to let the time interval

used for smart charging depend on the variance (standard deviation) in stop and start

times. A drawback of the above method is that cyclic daily or weekly variations are

not taken into account.

The method was included by the Thesis Author in the demonstration software de-

veloped in EDISON. The following figure 4.2 shows a part of a fleet operator interface

where the lines in the top portion “EV charging availability” illustrate how the fleet

operator transitions from using a static guess/profile of night availability to using a pre-

diction based on exponential smoothing. The upper line is the “measured availability”

while the dashed line underneath is the “predicted availability”.

Figure 4.2: Implementation of predictions - Illustration of EV charging availabilityprediction by a fleet operator as implemented in EDISON software.

The above represents a first approach to making predictions on night plug in avail-

ability and its active use by fleet operator aggregation software.

46


4.3.2 Early Investigation of Real Driving Data

The Thesis Author has investigate real EV driving patterns as part of the Danish Test-

An-EV programme. The programme, orchestrated by the electric mobility operator

(EMO), Clever, includes over 1600 Danish households over a three year period with

a fleet of 184 EVs. The project is supported by the Danish Energy Agency and the

Danish Transport Authority. The data has been stripped of personal information to

protect the identity of the participants.

In the study, groups of eight to ten EVs are distributed to cities in Denmark, here

the groups participate in a number of ”rounds”, each with a duration of approximately

3 months. In every round, each EV is given to a household. The households all receive,

and hand back, the EVs at the same time. For this study, six cities have been chosen,

each with two rounds as illustrated by figure 4.3.

Figure 4.3: Full sample of vehicles - Approximately 50 EVs and 100 families drivingin six different cities throughout the winter, spring and summer months of 2012.

The above is the data used for the plug in investigation.

Figure 4.4 visualises the charging periods for a single vehicle from the Test-An-EV

dataset during spring 2012. Notice that charging periods lies in the very beginning of

each plug in period signifying that, what is commonly known as ”instant” or ”dumb”-

charging, is used.

47

4. EV OWNER

Figure 4.4: Plug in periods - Visualisation of plug in periods for a single vehicle.

It has already been argued that the most important parameters for a plug in, in

regards to controlled charging, are the duration, predictability and the energy demand

for the next trip. These parameters are now investigated based on the recorded plug

ins.

First, however, the sample is filtered to only include ”workday night charge plug

ins”, i.e. a plug in starting after noon and terminating the following day before noon

on a workday (monday to friday). Some of the vehicles have been excluded due to

inactivity and/or lack of data.

Since a plug out event is not recorded in the data, it is assumed that the EV will

remain plugged in until the next drive if it was plugged in at the end of the previous

drive.

The averages for the full sample (illustrated in figure 4.3.) is listed in table 4.1.

It can be seen from the numbers that the charge period is only roughly one third

of the plug in period meaning that there, based on the average, is a lot of flexibility

available in when to charge the battery. This is even though the batteries are charged

to 100 percent and it typically takes considerably longer charging the last 5-10 percent

48


Parameter Mean value Standard Deviation

Plug in period duration 12:43:03 41 (Min)

Charge period duration 04:00:12 17 (Min)

Plug in period start SOC 49 4 (Percent)

Plug in period end SOC 100 2 (Percent)

Plug in period start 19:10:28 39 (Min)

Plug in period end 07:53:32 29 (Min)

Table 4.1: Plug in period averages - full set.

due to the BMS and its management of battery lifetime.

It can also be seen that the average State-Of-Charge (SOC) is relatively high at

plug in, and that the plug out (Could be when EV owner leaves for work) is subject

to less variance than the plug in (Could be when EV owner returns to his/her house).

Such ”fleet” information is valuable to an fleet operator for accessing the potential for

controlled charging.

Subsets of the above data have been put to use in a few investigations described in

the following sections. In these investigations, the sample sizes have been kept relatively

small, this is due to the fact that the 2012 data, especially from late 2012, still needs

to be thoroughly tested for completeness and correctness before it can be included in

a larger study.

4.3.3 Plug in Patterns: Seasonal Changes

This study is an investigation of possible changes in plug in behavior due to seasonal

changes. To this aim two samples are compared from the month of February and June

2012 - see figure 4.5.

Figure 4.5: Winter vs. summer month - Comparison between the plug in periodparameters of the month of February and June 2012.

A study have been conducted to compare the plug in periods of ten EVs driving in

February (February-set) with ten EVs used in June (June-set). Both the February-set

49

4. EV OWNER

Parameter February -Mean

February -SD

June - Mean June - SD

Plug in period duration 12:32:30 66 (Min) 11:38:20 85 (Min)

Charge period duration 04:06:02 31 (Min) 04:12:35 20 (Min)

Plug in period start SOC 50 7 (Percent) 46 5 (Percent)

Plug in period end SOC 100 1 (Percent) 99 7 (Percent)

Plug in period start 19:40:15 64 (Min) 20:06:29 56 (Min)

Plug in period end 08:12:45 22 (Min) 07:44:49 72 (Min)

Table 4.2: Plug in period seasonal averages - February and June set.

and June-set of EVs are equally distributed between the same two cities. The results

can be seen in table 4.2.

From an fleet operator point of view the February-set vehicles seems to have greater

flexibility in that they are plugged in for longer durations. A possible reason could be

an altered behavior due to changes in hours of daylight (People might get up later and

arrive home earlier during winter).

It is however dangerous to claim such causality based on the limited amount of

samples used in the data. Even if having the same distribution of vehicles betweens

cities in each set may remove some geographical bias, it is still different families driving

the vehicles and such bias can only be compensated for by increasing the sample size.

The above should therefore be supplemented by other larger investigations.

4.3.4 Plug in Patterns: Familiarization

Another interesting study question is whether there is a change in plug in behavior

when comparing the initial part of the test period with the last.

If, for instance, an EV owner plugs in less often due to increased “range confidence”,

it will result both in less availability and, when the EV is available, a higher energy

demand. The result is a significant reduction in flexibility as expressed in the flexibility

function described in the introduction 4.1.

This study is illustrated in figure 4.6 and includes 12 EV test families distributed

equally between four cities. The first, approximately 90 days, of a test period constitutes

the initial period while the last 90 days represents the mature period. The result is

presented in table 4.3.

50


Figure 4.6: Initial vs. mature use - Comparison between the first half and last halfof the test periods during the winter/early spring months of 2012.

Parameter Initial -Mean

Initial - SD Mature -Mean

Mature -SD

Plug in period duration 13:03:43 52 (Min) 12:41:44 32 (Min)

Charge period duration 04:06:08 23 (Min) 04:03:55 15 (Min)

Plug in period start SOC 48 4 (Percent) 49 4 (Percent)

Plug in period end SOC 99 4 (Percent) 100 0 (Percent)

Plug in period start 19:04:47 47 (Min) 19:21:25 45 (Min)

Plug in period end 08:08:30 37 (Min) 08:03:09 27 (Min)

Table 4.3: Plug in period familiarization averages - Initial and mature set.

51

4. EV OWNER

Based on this sample there seems to be little difference between the initial and

mature part of the test periods. Also the amount of plug ins are relatively stable

throughout the two sets. The EVs seems, however, to be connected a bit longer (13:03

vs 12:41) in the initial period and for these 12 vehicles and test families, this results in

the fleet having greater flexibility 4.1 and perhaps, value, for the fleet operator.

The relatively small sample size and time period in consideration makes the result

more sensitive to vacation periods. Again, it is the aim to repeat the study when more

data have become available.

4.3.5 Plug in Variance Case and Relation to Smart Charging

This section will concentrate on a single vehicle to illustrate how charging flexibility may

be used for different purposes. An EV driving in northern Jutland have been selected.

The EV has a total of 55 night workday plug ins over the three month trail period

signifying a very frequent and consistent use. Figure 4.7 shows the SOC distribution

at the start of plug in.

Figure 4.7: SOC distribution for case EV - Distribution of 55 SOC measurementswith a mean of 39 (SD 19).

The vehicle is used for long daily commutes which is reflected in the relatively low

average plug in SOC.

For the same vehicle, the cumulative distribution of plug in and plug out times are

illustrated. For the sake of the illustration the distribution of the plug in times are

”inverted” so that the ”valley” in the graph represents a high probability of the EV

being available to the grid. The variance of the SOC and Plug in/out distributions

matters since a high standard deviation may can result in a higher risk for controlled

charging. I.e. the fleet operator can underestimate the energy need or the amount of

time available. It may be necessary to consider a trade-off between risk and benefit.

52

4.4 Sub-conclusion

For this case it is riskier to postpone charging to the early morning hours between

5:30 and 6:45. The ”risk-area” would increase if the distribution were more negatively

skewed i.e. left tailed. If a specific ”expected time of plug out” is chosen (i.e. 6:20) the

fleet operator can tell the owner that this, based on the plug out distribution, would

be appropriate roughly 90 percent of the time. The EV owner should than accept the

slight risk of leaving with less energy than planned when plug out happens earlier than

expected.

The only reason the fleet operator and owner would take such a risk is if the benefit

of charging is high in the same hours where there is a small chance a plug out may

occur. Decision making under uncertainty is left to chapter seven: ”Optimization”

later in this thesis.

This section concludes by showing some of the possible drivers for controlled charg-

ing. Data for all sources are collected for the same period (April to June) and part of

Denmark (Western Denmark) as was the case for the family testing the EV.

Of the sources both wind and market prices seems to favour night charging. For

wind, the average production is a bit lower during night but remains more stable than

the demand which falls considerably. It is important to add that looking at the dynamic

behavior in a sub-hour time-scale, wind production can exceed demand at several points

in time. The graph only hints to the possibility of charging based on wind energy - not

on power services such as regulation/ancillary services.

For a house-mounted PV however, there is little synergy to be found. The short

period during the morning (around 6:20) where energy potentially could be drawn from

PV to EV represents a high risk (for the EV leaving) small benefit (only few Watts of

power) scenario.

In chapter seven ”Optimization”, the price of energy is used as motivation for

controlled charging.

4.4 Sub-conclusion



53

4. EV OWNER

Figure 4.8: Plug in/out cumulative distributions and input for value functions- Visualisation of availability compared to the sources that may motivate controlled nightcharging.

54

4.4 Sub-conclusion


In a study based on ICE vehicles the following have been done:

• An approach to simple plug in period prediction based on exponential smoothing

have been proposed.

• The above method have been implemented in demonstration software for EV fleet

management as part of the EDISON project.

Next, it is demonstrated how real EV data may be used to investigate plug in

behavior. This includes:

• A description of the main parameters that determines plug in periods usefulness

for controlled charging.

• Early indications on the effects that EV familiarisation and different seasons may

have on plug in behavior.

• A specific EV is used to illustrate plug in availability compared to different value-

sources for controlled charging.

The author hopes that such investigations can help uncover the role and impact

that EV owner plug in behavior has in regards to intelligent EV utilization concepts.


The control systems managing charging, should know the plug in patterns of the EV

owner as well as possible. This can both increase convenience by reducing the need

of EV owner intervention in smart charging management, and at the same time make

sure that the flexibility of the vehicle is used optimally without violating the driving

needs of the EV owner.

Investigations on the Test-An-EV data indicates that there is a lot of charging

flexibility available for the families included in the test programme (plug in period

duration compared to energy demand). For a specific sample investigated, there seemed

to be more flexibility available during a winter month compared to a summer month.

EV driving familiarization does not seem to have a great impact on flexibility based on

55

4. EV OWNER

the small sample investigated. The studies should, however, be validated using a larger

sample to reduce the sampling error.

The fleet operator will have to describe the risk/benefit trade-off of controlled charg-

ing to the EV owner, i.e., the EV owner must accept that the fleet operator may mis-

calculate the charging flexibility 1 out of X times. Based on the plug out probability

distribution the fleet operator may, for instance, choose an expected plug out time that

in three percent of the cases will be later than the actual plug out. If the owner then

plugs out earlier than expected, and the fleet operator had planned charging after that

point in time, the EV may leave with less energy than intended. A risk described by a

percentage may be easier to communicate to the EV owner than a more sophisticated

expression of the risk/benefit trade-off.

There are numerous ways in which EV owner behavior can be modelled and pre-

dicted. These can, however, be kept relatively simple if night charging is the only

period considered and mechanisms such as ”instant charging” and ”minimum range”

features are used to handle outliers (e.g. extraordinary mid-night driving needs).

An important issue in relation to data collection and prediction of behavioral pat-

terns, is the protection of EV owner privacy. This have to be dealt with pro-actively

by the fleet operator.

56

5

Control Architecture

This chapter deals with the control architecture as part of the “Supporting informatics”

investigations of this thesis. The following publications constitute the core contributions



C Smartly Charging the Electric Vehicle Fleet.

The publications will be referred to by their letter (B,C) in the following.

5.1 Introduction

When the purpose of controlled charging and the requirements and impacts of key

stakeholders have been established, the next task will be to build a suitable control ar-

chitecture. Choosing the control architecture comes down to deciding what entities will

have control over charging, where physically to place the necessary control systems and

software logic and finally to establish the, possibly hierarchical, relationship between

such systems.

The same aim, for instance smart charging, can be achieved by several different

control architectures. There may, however, be differences in what is most practical

and efficient to implement. This chapter is the first within the supporting informatics

part of this thesis since the choice of control architecture will shape the form of the

communication solutions and optimization models to be used. Basically there are two

57

5. CONTROL ARCHITECTURE

high-level architectures to choose from - Centralized or distributed. These options are

shown in figure 5.1

Figure 5.1: Centralized and distributed architecture - The high-level architecturaloptions for EV integration.

A centralized architecture is generally characterized by having the control and de-

cision making delegated to a common coordinator and letting each Distributed Energy

Resource (DER) be directly controlled by this entity. In contrast, a distributed design

means that each DER will act as an independent, intelligent, and autonomous agent

that responds according to a set of incentives and information. The above is a coarse-

grained distinction and several flavors and variations exist. The centralized approach

maps nicely to the Virtual Power Plant (VPP) concept while the distributed approach

can be related to the field of agent theory.

Another important aspect is how a resource, in this case an Electric Vehicle (EV),

is described to the surrounding world. Another useful distinction can be introduced:

Structural decomposition Each detailed component, all measurements and the ex-

act operational state of a unit is communicated to external entities with the

proper permissions to receive such information.

Functional decomposition Internal details is hidden and the resource is rather de-

scribed by the functional services that the unit may provide.

The former category fits well with the approach in classical SCADA-like automation

approaches as that of IEC 61850 which will be described in the next chapter. The latter

58


fits with the more autonomous, lowly coupled, approach followed in the service oriented

concept seen within contemporary internet use. A final consideration covered by this

chapter is what high-level functionalities that should be implemented in the control

system after a control architecture has been chosen. In this thesis the centralized control

architecture is chosen and the “Thesis work and contributions” section describes why.


One of the largest and first European projects to thoroughly explore the use of the VPP

concept was the FENIX project (2). This project introduced a distinction between a

commercial VPP (CVPP) and a technical VPP (TVPP). The commercial version would

act on the power market to meet a financial goal. In contrast, the technical version

would be used to provide technical services to the grid.

Another thorough study of the VPP concept has been done by Shi in (4). Shi

has focused on another type of the VPP concept - the generic virtual power plant

(GVPP). The reason for a VPP being generic is that it would be able to handle and

aggregate various types of DERs. As will be done for the EDISON VPP/fleet operator

in this chapter, Shi has suggested the generic, high-level functions that would need to

be implemented to support the operation of the aggregator. See figure 5.2

For reference, it can be interesting to investigate the control architecture chosen by

a set of EV integration projects. This was investigated in core publication B and is

illustrated in figure 5.3.

As can be seen, both the e-mobility project Berlin and the University Of Delaware

(UD) Vehicle-To-Grid (V2G) programme uses partly-centralized approaches. The e-

mobility project, like what is the case for UD V2G, lets the EV itself implement software

that will dictate the degree to which the EV will participate in intelligent EV utilization

concepts.

The UD V2G programme depends on the area of agent theory by defining local

software agents (JADE framework) for each device in the system (39) - e.g. an EVSE

agent, an EV agent and an aggregator agent. Each of these have distinct and unique

objectives and the communication between them is “negotiation-like” in nature. In-

terestingly, since the implementation is aimed at current markets designs in the PJM

area, a number of the responsibilities still have to be left to the aggregator and thus the

59


Figure 5.2: Function-based design for a GVPP - A generic and reusable model thatgeneralizes the objective of the VPPs via a function-based approach. Source: Shi You (4)

Figure 5.3: Comparison of architectures - Architectures chosen in a set of state-of-theart electric vehicle demonstration projects

60


solution can be considered a hybrid between the centralized and de-centralized design,

i.e., the individual EV agent will be given responsibility of making sure that driving

demands are meet - but will depend on the aggregator agent in connecting and adhering

to the power market.


This section presents the architecture chosen as part of the EDISON project and how

it was used to implement a EDISON VPP back-end system, used to demonstrate the

operation of a fleet operator.

5.3.1 System Architecture of EDISON

In the EDISON project, the choice of control architecture for demonstration software

was left to Work Package 3 (WP3) in which the Thesis Author participated. Here the

choice became using a centralized design. The main reason being that this approach

best suits the current market models and available EV/EV Supply Equipment (EVSE)

technology.

The use of agents is projected to allow for a more flexible and dynamic forming of

ad-hoc groups of DERs for meeting a broad array of goals (39). It is, however, the

opinion of the Thesis Author and the other participants of EDISON WP3, that agent

technology may take more years to implement in the power system than is the case of

a centralized approach.

EDISON WP3 was inspired by the distinctions made in FENIX between a technical

and commercial VPP and therefore explored both options. In the end, the choice fell

on the commercial VPP since the market allows for providing technical services to the

power system while at the same time releasing a monetary benefit to stakeholders. This

also made the implementation easier as the interface to the market is better defined

than an interface directly to a system operator would be. The two possible approaches

are illustrated in figure 5.4.

Relating the choice of VPP control architecture in EDISON to the variations de-

scribed by FENIX and Shi, The EDISON VPP then becomes centralized, non-generic

(focused on the EV as a resource) and commercial (interface with the market). Again,

61


Figure 5.4: Architecture in EDISON - Approaches for interfacing with the powersystem either directly (Via DSO) or to the market (Via market players) Source: BernhardJansen, IBM, EDISON internal report

it can be argued that participating in present and future ancillary service market prod-

ucts (see market chapter) makes it is less necessary to distinguish between a technical

and commercial purpose of a VPP - both would be done at once.

5.3.2 High-level Functions

Having defined the architecture, the focus was then put on the functions needed in the

EDISON VPP used by the fleet operator. Figure 5.5 presents such functions.

This functionalities are presented in a book chapter as part of the core publication

C The figure conveys that the control system would have to handle the EV fleet as

an aggregated group when acting and optimizing toward market players, but will have

to take individual considerations into account when managing the charging behavior

of individual EVs. This observation has an impact on how to form the optimization

problem for the fleet. Another property of the figure is that functionality is divided

into three distinct groups. Namely ”data”, ”analytics”, and ”logic”:

Data This group stores previous market prices and fleet behavior on an aggregated

62


Figure 5.5: High-level centralized fleet operator functionalities - Functions at theindividual and aggregated level, subdivided into ’data’, ’analytics’ and ’logic’ )

63


level, enabling better forecasting and optimization for acting on the power market.

On an individual level, data is stored that describes the service level agreement

between the EDISON VPP and an EV owner, EV hardware specifications and

the EV owner’s plug in habits.

Analytics This group covers the mathematical computations necessary to support the

logic of the EDISON VPP. Forecasting relies on historical data to predict market

prices on an aggregated level which supports better bid strategies. Forecasting

also determines future individual EV usage patterns. Optimization is used to

minimize charging costs of the EVs on both the aggregated and individual level.

The individual optimization is limited by the constraints introduced by the dis-

tribution grid, EV specifications and EV owner energy requirements. On the

aggregated level, profit maximization can be done when acting on the regulating

and reserve markets.

Logic This group defines the main operational goals of the EDISON VPP, namely to

act on the power market to generate savings or revenue for itself and its clients,

and to intelligently manage the charging behavior of EVs through individually

tailored charging schedules.

The demonstration software developed as part of EDISON implements the above

functionalities.

5.4 Sub-conclusion




The architectural choice in EDISON has been inspired by architectures explored by

other research projects and academia. The Thesis Author have participated in building

an architecture that is:

• Centralized, and thus depends on an aggregating entity.

• Based on an commercial objective, trading on the market.

64

5.4 Sub-conclusion

• Focused on managing EVs, rather than being generic.

Having established the control architecture, the Thesis Author participated in building

a back-end system to demonstrate coordinated charging. To this end the follow was

done:

• Necessary functionalities were defined in terms of data, analytics and logic.

• The backend system was developed in code and was used in end-to-end tests in

EDISON.

• The developed back-end system facilitated tests of communication, prediction and

optimization as well as the EV and EVSE hardware included in the field trails.


The EDISON project have identified a centralized approach as the most suitable short-

term architecture for EV integration. As the power markets and power system evolves,

the attention may be turned to a more decentralized approach, e.g., using software

agents.

An interesting aspect of the placement of intelligence and control inside or outside

the EV itself, is the approach chosen by car OEMs. OEMs involved in early smart

charging schemes such as Nissan and the Daimler group have promoted solutions where

the EV is in charge of the charging process. This is not necessarily because the OEMs

are strong believers in agent technology but rather because they want to protect the

EV and its battery from external mismanagement and at the same time protect the

business potential of intelligent charging.

A conclusion drawn from the core publication B, regarding the architecture chosen

by different projects is that the placement of control logic and responsibilities between

EV, EVSE and aggregator, differs between the integration projects investigated. One

key question is how much intelligence will implemented at the EV. If a common archi-

tecture is not found, then standardization efforts must respect and manage the resulting

diversity.

65


66

6

Communication

This chapter deals with communication as part of the “Supporting informatics” inves-

tigations of this thesis. The following publications constitute the core contributions



C Smartly Charging the Electric Vehicle Fleet.

F Facilitating a generic communication interface to distributed energy resources - Map-

ping IEC 61850 to RESTful Services.

The publications will be referred to by their letter (B,C,F) in the following.

6.1 Introduction

Having established a control architecture, the next question is how to transfer the

necessary pieces of information between the control systems in a reliable, robust, secure,

and standardized way.

Typically a communication standard consists of the following components:

Data model The formalized description of the structuring and formatting of data.

Protocol mapping The protocols and technologies that are used to transfer the data

across a network.

67

6. COMMUNICATION

Smart grid informatics research has already presented a host of possible communi-

cation solutions. Often smart grid integration sub-areas have their own set of communi-

cation solutions. Most types of Distributed Energy Resources (DER), for instance, have

their own set of standards. Take for instance the IEC 61400-25 communication interface

standard for wind power plants, home automation standards such as Open Building

Information Xchange (oBIX), AS 4755.3.1-2008 (Air Condition demand response), and

the work of the OPEN meter network to standardize smart meter communication. Such

standardization efforts are also found within Electric Vehicle (EV) integration.

It can be useful to divide communication standards into a few categories:

Firstly, communication solutions can either be said to be generic or application

specific in nature depending on how many scenarios and devices they are designed to

cover. New smart grid application areas can either be addressed by extending existing

standards or by proposing new ones.

Secondly, as touched upon in chapter five, “Control architecture”, communication

standards can either use a structural of functional decomposition of resources, i.e.,

describe a device in terms of its exact operational parameters and physical components

or in terms of the high-level functions and services that it can provide.

In relation to the categories mentioned above, IEC 61850 can be called a generic

standard using structural decomposition. It also represent a relatively mature standard

and it therefore seems reasonable to investigate its potential in EV integration before

suggesting new solutions.


The use of the IEC 61850 standard for substation automation is thoroughly investigated

and documented (40, 41). Such publications describe the potential that IEC 61850 has

in providing monitoring and control capability of in-field equipment in an interoperable

and standardized fashion. Also, the extension of IEC 61850 to applications beyond

substations have also been thoroughly researched (42, 43).

As mentioned in the introduction, most communication standards can be described

in terms of their data models and protocol mappings.

An investigation regarding the IEC 61850 protocol mapping can be found in (44)

by Schmutzler. Here, Schmutzler demonstrates how Devices Profile for Web Services

68


(DPWS) can be mapped to IEC 61850. DPWS describes how the web service concept,

and the WS* technologies behind it, can be used towards embedded devices. Schmutzler

have thereby demonstrated how a popular and feature-rich web technology can be

applied to IEC 61850 to increase interoperability.

Other research also cover the data description of IEC 61850. In (45), Sui suggests

using OPC UA instead of MMS and supplementing the data model of IEC 61850 with

OPC UA information modelling. The purpose of applying OPC UA is to promotes

DERs as Service Oriented Architecture (SOA) ready devices.

The possibility of extending IEC 61850 standard to many different applications, us-

ing different contemporary communication technologies, increases its potential towards

EV integration.

In B, a comparison is made between the communication technology solutions chosen

by three projects as seen in figure 6.1.

Figure 6.1: Project comparison of communication technologies - Comparison ofthe communication technologies chosen by three EV integration projects.

The projects all use the internet and the associated TCP/IP stack on the net-

work/transport layer as well as Power Line Communication (PLC) on the physical

layer (between EV and EVSE). While the projects are aligned on the lower parts of the

OSI stack, they are very divided on the application level protocols. The solutions span

from the structural decomposition used by IEC 61850 to the use of agent software.

The EDISON project is the only of the investigated projects to use a standardized

communication protocol (IEC 61850).

If IEC 61850 is used between fleet operator and EVSE, as described in the following

section, there is still a need of communication solutions between EVSE and EV.

69

6. COMMUNICATION

Some of the options available for EV to EVSE communication have been investi-

gated in a special project (46) supervised by the Thesis Author. In the project report,

Johansen compared the use of Negative Side Signaling (NSS) and PLC for the physical

communication layer. The report points to PLC as being the most likely candidate

for use in future EV integration. For PLC the main choice is between using wide- or

narrow band technologies. HomePlug GP is described as a wideband solution with

a data rate of up to 10Mbps. There are some technical (limited range, no support

for both DC and AC) and regulatory (reserved frequency bands) limitations on using

wideband technologies. One of the main advantages of wideband, however, is a higher

data rate compared to narrowband. The rates given by narrowband is sufficient for

controlled charging-related communication but may not support additional communi-

cation applications, e.g., multimedia. The report also investigates the use of a tentative

version of the ISO/IEC 15118-2 standard for the upper communication layers. The use

of IEC 15118 and PLC, as part of the EDISON project, is also touched upon in C. The

emphasis in the thesis and the following sections is on communication between EVSE

and fleet operator.


This section will describe the Thesis Authors experience with the IEC 61850 standard

for the application of EV integration.

6.3.1 Research on the IEC 61850

Figure 6.2 illustrates the main components of IEC 61850 based on the structural de-

composition of a Micro Combined Heat and Power Unit (CHP).

The figure includes the reusable “building blocks” of the data model (logical nodes,

data sets etc.) and the suggested communication protocol mapping (Manufacturing

Message Specification (MMS) on top of the TCP/IP stack using Abstract Communi-

cation Service Interface (ASCI) definitions). For a description of the standard refer to

core publications C and F.

While working with the standard, the Thesis Author together with co-authors in F

have suggested extensions and additions to both the protocol mapping and data model.

70


Figure 6.2: The 61850 standard - Main concept and components of the 61850 standard- data model and protocol mapping, source: Anders Bro Pedersen(47).

6.3.1.1 Using REST Services

As mentioned above, the IEC 61850 standard initially pointed only to MMS as the

recommended application level protocol. While MMS represents a valid choice for IEC

61850 implementation, there are certainly other approaches which can and should be

explored.

An obvious candidate is the use of web service technologies, which have seen use

in a broad set of domains to promote SOA. Web services represents a well-understood,

open, interoperable and feature-rich framework to grant access to information services

on the web. Web services have, for instance, seen early support in standards such as

61400-25 for wind power plants.

Web services are often associated with the Simple Object Access Protocol (SOAP).

The addition of SOAP based extensions to web services provides a rich list of features

and extensions. These features and extensions are captured in a set of specifications

referred to as “WS-*”. Instead of focusing on SOAP-based services, however, another

alternative exists; REpresentational State Transfer (REST) services. REST services

(48) is considered a simpler and more web-centric approach to web services that uses the

basic features and methods (GET, POST, PUT and DELETE) of the HTTP protocol.

71

6. COMMUNICATION

In REST a services is accessed as a resource and is “fetched” as you would an HTML

page or image, via a URL.

The IEC 61850 reference paths resemble a fully-qualified file-name notation which

can be mapped directly to a URL as visualized in figure 6.3.

Figure 6.3: Mapping to URL path - REST URL to access current charging power ofan EV source: Anders Bro Pedersen(47).

REST services are implemented by following a set of principles (Use of HTTP,

everything is a resource, etc.) and is not, as SOAP services, a standard.

6.3.1.2 Data Model

To move the IEC 61850 standard beyond the substation environment, an extension

called IEC 61850-7-420 was developed to address the integration of DERs into the power

system. A new set of logical nodes, one of the main building blocks of the IEC 61850

standard, was introduced to cover the components, measurements and functionalities

of such units.

An interesting aspect of using IEC 61850-7-420 was to see how useful these nodes

were at representing an EV. The results of the study were that the logical nodes in 7-420

not fully represented the needed values, especially for EV batteries, such as SOC and

battery capacity. As a result of this work, two new logical nodes were suggested rep-

resenting an EVSE (ZCHS) and an EV (ZCEV). These additions have been suggested

toward the IEC standardization body for consideration.

72


The logical nodes from IEC 61850-7-420 which were used in the implementation,

as well as the new ZCEV and ZCHS nodes (Marked with “61850 Extension”), are

illustrated in figure 6.3

Figure 6.4: Logical nodes used in implementation - List of logical nodes used torepresent the EVSE and EV as part of the data model of IEC 61850 source: Anders BroPedersen.

6.3.1.3 Test of IEC 61850/REST Implementation

An IEC compliant server have been developed as described in (47) and have been tested

in a series of demonstrations towards electric vehicles as part of the EDISON project.

The full EDISON communication setup is illustrated by figure 6.5.

The communication setup have been tested towards different brands of EVs in

EDISON. Most vehicles were monitored in a read-only fashion while a single vehicle

was controlled via IEC 61850 based schedules (”DSCH - Energy and/or Ancillary Ser-

vices Schedule” and “DSCC - Energy and/or Ancillary Services Schedule Control”). A

screenshot from a demonstration with such vehicles can be seen in figure 6.6.

73

6. COMMUNICATION

Figure 6.5: Communication setup in EDISON - Communication standards used toconnect the EV, EVSE and fleet operator.

Figure 6.6: Test of IEC 61850 communication - Use of IEC 61850 towards a vehiclefleet - data such as status and SOC is displayed in a fleet operator interface.

74

6.4 Sub-conclusion

6.4 Sub-conclusion




The Thesis Author have contributed to the following activities relating to communica-

tion solutions for EV integration:

• IEC 61850 have been evaluated for the application of EV Integration.

• The use of REST services as a protocol mapping for IEC 61850 have been inves-

tigated.

• A list of logical node extensions have been suggested as an addition to IEC 61850-

7-420 to better represent EVs and EVSEs. They have been sent to IEC WG17

for possible inclusion in an updated version of IEC 61850-7-420.

• The suggested communication setup have been tested in a set of EDISON-related

demonstrations.

The communication qualities, focused upon throughout this chapter, have been

simplicity and interoperability. Other qualities, such as security and latency, are of

course also important and are discussed in the following section.


Security is an important topic, both for EV integration and for smart grids in general.

A lot of solutions for providing confidentiality, integrity and availability can be pro-

vided by “of-the-shelf” technologies already tested in other applications. For the IEC

61850/REST implementation the use of the Transport Layer Security (TLS) as a cryp-

tographic protocol can be used to ensure confidentiality and integrity. This protocol

have seen applications in security-critical applications such as web banking.

If using SOAP based web services on top of the IEC 61850 standard, Web Service

Security (WS-security) can be used. WS-security describes the use of signed SOAP

messages for integrity, encryption in SOAP messages for confidentiality and the use of

security tokens for guaranteeing sender identity.

75

6. COMMUNICATION

If MMS is used as the application layer protocol, on top of IEC 61850, a security

profile is recommended in the “IEC 62351-4 Security for any profiles including MMS”.

The IEC 62351 standard was made specifically to define security for a number of IEC

Technical Committee 57 communication standards - including IEC 61850. IEC 62351

also points at TLS as a protocol for providing security.

In general, it is recommended to align EV communication security with smart grid

related security standardization work from standards organisations such as IEC (e.g.

IEC 62351) and NIST (e.g. “Smart Grid Cyber Security guidelines”).

Latency in communication depends in part on the inevitable delay caused by the

use of the internet and the associated routing of data over the nodes in a huge meshed

network. Another part is the overhead, in term of size, that the protocols used add to

the data packages.

Theoretically, the use of REST should reduce the amount of data to be sent since

the overhead of using SOAP is avoided. It is, however, unknown if this makes a mea-

surable difference. Latency in communication have not been a topic of investigation in

this thesis, but the following observation can be made; In demonstrations performed

over the internet with the IEC 61850/REST stack, data was received within a sub-

second time scale. If the application domain is ancillary service market integration

(”Market integration” chapter) and the shortest required response time here is 30 sec-

onds (”Primary reserve” and “Frequency controlled disturbance reserve”) then latency

is not a problem. In some of the most time-critical intelligent EV integration concepts

proposed, the EV would react directly on local measurements of net-frequency, and

communication latency would not matter.

There are several other communication issues that should also be considered, among

them are scalability, accessibility and firewall traversal. For the latter issues a solution

using SIP have been suggested by Jansen in (49). SIP provides a means for creating a

point-to-point connection between clients not accessible through a global fixed IP. For

firewall traversal, Session Traversal Utilities for NAT (STUN) and the Traversal Using

Relay NAT (TURN) is suggested. This approach is also described in C.

Looking to the future, it is likely that new standards covering EV communication

will emerge. This chapter have already touched upon the ISO/IEC 15118 which defines

how EV to EVSE communication can be implemented. A protocol possibly rivalling

76

6.4 Sub-conclusion

IEC 61850 for EV use, is the Open Charge Point Protocol (OCPP). Flavours of OCPP

is already supported by a series of EVSE vendors.

Communication standards should support as many intelligent EV utilization con-

cepts as possible. For the IEC 61850 implementation described in this chapter this

meant adding logical nodes to describe battery SOC and the use of energy Schedules

to control charging - this allowed for the concept of smart charging to be realised.

It will be interesting to see if future smart grid standards will be as broad and

generic as IEC 61850 or more application specific like IEC 15118 and OCPP. It will

also be interesting to see whether structural or functional decompositions will be the

norm. It can be argued that the latter is a better match with the SOA paradigm often

pursued in modern web implementations.

77

6. COMMUNICATION

78

7

Optimization

This chapter deals with charging optimization as part of the “Supporting informatics”

investigations of this thesis. The following publication constitute the core contributions



EDISON project.

The publication will be referred to by the letter (E) in the following.

In addition to the existing publication, new material, mentioned in the sections on

markets and stochastic optimization, are presented in this chapter. This work is part of

a study attempting to combine real Electric Vehicle (EV) data (new material in chapter

four “EV owner”) with stochastic programming (this chapter).

7.1 Introduction

Mathematical optimization is used to make the best possible decisions given a well

defined objective and a specific domain. Applied to this domain (intelligent EV inte-

gration), objectives can be dictated by the market (chapter two) and the constraints

by the distribution grid (chapter three) and the EV owner (chapter four). The transfer

of EV owner, grid and market data must be supported by the used control architecture

(chapter five) and communication technologies (chapter six).

An important issue is the information on which the optimization will be based. It is

necessary to understand what information will be available at each operational stage,

79

7. OPTIMIZATION

the possible uncertainty of such information and the information “certainty gain” as

time progresses.

Chapter three “Grid integration” described how grid capacity information may be

made available to the fleet operator and chapter four: “EV owner” described how

plug in behavior may be predicted. The latter chapter links directly to this one, by

presenting the data which can be used for scenario generation.

The following figure illustrates the approach used in this chapter where predicted

or known information is used for defining a suitable charging schedule for an EV.

This process could be repeated as new information is made available throughout the

operational stage.

Figure 7.1: Optimization approach - Approach for calculating a cost minimized charg-ing schedule based on known and uncertain information.

The figure also shows how the schedule is used on an individual (towards EV) and

aggregated level (towards market). The distinction between operating on an individual

and aggregated level was introduced in chapter five “Control architecture”. Using the

approach in this chapter would mean that the fleet operator would first make individual

schedules and then “sum them up” to arrive a the aggregated bids and actions on the

market.

The fleet operator can not avoid individual charging management, due to individual

hard constrains, but a possible refinement of the above approach, not explored here,

would be to consider the EVs as a fleet until the operational stage.

Using fleet averages on demand and availability may be sufficient for acting on the

market since there will be a “smoothing effect” when enough EVs are part of the fleet.

This approach would also be computationally less heavy.

80


Making individual EV schedules during the planning phase, might, on the other

hand, yield the most accurate market bids. These issues of individual vs. aggregated

portfolio management is briefly discussed in the last section of the chapter, but mostly

left to future research.

This chapter will focus on charging cost minimization for a single vehicle considering

the access to different markets and uncertainty in plug in, demand and grid capacity.

The mathematical model which will be presented here has been implemented and

tested using ”General Algebraic Modeling System” (GAMS) software.


In a first wave of research within EV fleet optimization, different combinations of opti-

mization objectives and constraint have been investigated. Two commonly investigated

objectives are either the minimization of charging costs or the maximization of the num-

ber of EVs the distribution grid can support.

From the former category numerous publications have demonstrated the use of

linear optimization for controlled charging. In the paper (50) the authors demonstrates

the minimization of charging cost using two different approaches for including battery

considerations. One using a quadratic approximation of battery behavior and another

using a linear approximation. The authors found that linear approximation sufficiently

represents the battery and is much faster to compute than the quadratic approach. In

(51), Sundstroem, Corradi and Binding introduces the use of time and energy margins

to compensate for prediction errors in plug in duration and energy demand. This is a

simple approach for handling stochastic EV owner behavior. In (52) Sundstroem and

Binding propose an optimal charging method where the fleet operator is aware of grid

constraints. Here, a cost-minimized charging schedule is validated by building a flow

network and solving the maximum flow problem.

The above papers have demonstrated that optimization methods can reduce the

energy cost for the EV owner and fleet operator while considering several different

constraints.

Concerning the objective of maximizing EV grid integration, a publication by Lopes

in (53) demonstrates that smart charging, specifically aimed at maximising EV pene-

tration, can successfully increase the number of EVs a distribution grid can support.

81

7. OPTIMIZATION

Symbol Unit Definition

cj kW Charging power at level j (only used by binary model)

Xtj 0/1 Binary decision variable indicating whether to charge at time intervalt at power level j (only used by binary model)

dt hour Duration of time interval t

pt Euro/kWh Energy price at time interval t

ct kW Charging power at time interval t

Aneg Euro Total cost of buying energy on adjustment market

Apos Euro Total income from selling energy on adjustment market

Bneg Euro Total cost of negative deviations on balancing market

Bpos Euro Total cost of positive deviations on balancing market

PAneg Euro/kWh Purchase price of energy on adjustment market

PApos Euro/kWh Selling price of energy on adjustment market

PBneg Euro/kWh Energy price of negative deviations on balancing market

PBpos Euro/kWh Energy price of positive deviations on balancing market

Dw kWh Energy needed for next drive in demand scenario w

Edefv kWh Deficit of energy caused by a late plug in, in plug in scenario v

Esurvw kWh Surplus of energy that can be sold in adjustment market in plug in

scenario v and demand scenario w

Kvw kWh Purchase of extra energy in adjustment market in plug in scenario vand demand scenario w

F defvo kWh The energy deficit created by grid congestion

Table 7.1: Symbol overview - Symbols used on the following pages

As more considerations are combined from the market, EV owner and grid, the

optimization methods will increase in complexity. This means that they will cover

more constraints, multiple objectives and should also deal with uncertainty.

An example of a publication that seeks to bring together several objectives and

constraints is (32), again by Binding and Sundstroem. The publication demonstrates an

optimization problem where interactions between a Charging Service Provider (CSP),

a Distribution System Operator and a Retailer seeks to both follow system-wide energy

prices while still adhering to grid constraints.


7.3.1 Cost Minimization

A first approach to optimization was suggested in core publication E. The approach

uses binary programming to decide in which time periods to place charging, based on

82


an energy demand and known prices. The availability time periods is predicted using

exponential smoothing. The objective function is defined as follows:

MinT∑

t=1

M∑

j=1

cjdtptXtj (7.1)

Where pt is the cost of energy at time interval t = 1, ..., T . The length of each time

interval is dt and cj is the charging power used from a set of available levels j = 1, ...,M.

In the objective function Xtj is used as the decision variable specifying whether

charging should be carried out at time t with charging level j. The function is then

subject to the following constraints:

M∑

j=1

X1j ≤ 1, ...,M∑

j=1

XTj ≤ 1 (7.2)

M∑

j=1

T∑

t=1

cjdtXtj ≥ D (7.3)

The first of the constraints 7.2 states that in all periods, charging can only be

carried out using one of the available charging levels. The second constraint 7.3 is used

to satisfy an energy demand D for the next trip. It could be possible to include grid

constraints in terms of charging limits using a matrix c ∈ RT×M+ of possible charging

powers where each row would be a timeslot and each column a supported charging

power.

The above represents a first approach to charging optimization using a deterministic

model.

7.3.2 Introducing Markets

In this section the above optimization problem is extended to include decision stages

where the fleet operator can use markets to meet the energy demand of an EV, while

minimizing the cost of charging. The following markets are considered:

Day-ahead market Market in which the fleet operator bids on energy before the day

of operation.

83

7. OPTIMIZATION

Intra-day adjustment market Intra-day market that allows the fleet operator to

compensate for positive or negative deviations after the day-ahead market has

closed.

The model will include two stages - one for each of the above markets. First the

fleet operator will attempt to meet the demand on the day-ahead market based on

the predictions on EV owner plug in time, demand and grid capacity. In the second

stage, when the actual plug in occurs, the exact demand and plug in time will be known

(certainty gain). The fleet operator must then use the adjustment market to compensate

for any energy surplus or deficit caused by the uncertainty. A final market - the

balancing market - is also included as a way to settle real-time imbalances introduced

by the fleet operator. This is not modelled as a stage since it is assumed that there is

no recourse for the kind of real-time events that necessitates the use of this market.

The stages are illustrated in figure 7.2. The picture uses the same terminology

(planning, scheduling, operation) as was introduced in core publication A regarding

the operational stages of a fleet operator.

Figure 7.2: Markets and stages - Use of markets in relation to charging cost minimiza-tion for EVs.

The above approach is not meant to accurately reflect the structure and mechanisms

for any specific market, but rather illustrate how the use of different sub-markets may

influence the optimization done by the fleet operator.

84


The markets are now introduced to the objective function defined in the previous

section. The problem is also changed from being binary (using Xtj) to linear.

Min

T∑

t=1

ctdtpt +Aneg −Apos +Bneg +Bpos (7.4)

In the above∑T

t=1 ctdtpt is the cost of energy purchased on the day ahead market

and ct, representing the charging power at time period t, is the decision variable.

Aneg and Apos represents the cost for negative and positive energy deviations which

can be handled through the adjustment market. Apos being negative denotes that

unused energy can be sold back to the market. It is assumed, however, that a ”Two-

sided penalty” scheme is used, i.e., buying and selling energy in the adjustment market

will always be less advantageous than in the day-ahead market.

Finally Bneg and Bpos is the cost for handling negative and positive deviations

through the balancing market. The prices are assumed to be considerably higher than

on the adjustment market and to represent a cost regardless of whether the deviation

is positive or negative.

Prices on all markets will be subject to uncertainty and the fleet operator will have

to rely on prediction methods in its planning. It has, however, been chosen to leave

this out of the model since the emphasis here is on EV owner and grid uncertainty.

Price scenarios can, however, be included in the model using the same approach as will

be described in the following sections.

Before introducing uncertainty, the fleet operator can avoid both the adjustment

and balancing markets. As will be seen in the following, however, uncertainty will make

the markets a necessary means for recourse in certain scenarios.

7.3.3 Introducing Uncertainty - EV Owner Behavior

The behavioral patterns of EV owners, in term of nightly plug ins, will be subject to

uncertainty as shown in chapter four “EV owner”. The two parameters which will be

considered here is the time of plug in and the energy demand. The energy demand is

the difference between the State-Of-Charge (SOC) seen at plug in and the target SOC

for the next trip.

To include uncertainty in the optimization, stochastic programming is applied.

Specifically, a two-stage stochastic linear program with recourse is introduced. To this

85

7. OPTIMIZATION

aim, scenarios for both energy demand and plug in time, are defined with an associated

probability for each scenario.

These scenarios are added to the objective function as follows:

MinW∑

w=1

πw

V∑

v=1

πv

T∑

t=1

ctdtpt +Anegwv −Apos

wv +Bposv (7.5)

In the above, the two summations∑W

w=1 and∑V

v=1 are added and the adjustment

and balancing market costs are altered to depend on the scenarios. It is, for this

case, assumed that handling negative deviations on the balancing market (Bneg) can

be avoided. For the first summation, w is a demand scenario with a probability πw.

For the second summation, v is a plug in time scenario with a probability πv. The

cost of each scenario is then calculated, weighted based on its probability and summed

together to give the expected average cost of the problem.

The cost of using the adjustment and balancing markets are calculated by the fol-

lowing equations. First the energy constraint defined in the first section 7.3 is redefined

as follows:

Dw ≤T∑

t=1

ctdt − Edefv +Kvw (7.6)

Here∑T

t=1 ctdt is purchased day ahead energy. Edefv is a possible deficit of energy

caused by plug in scenario v. Kvw is extra energy that would have to be purchased

depending on the demand w and plug in scenario v. Finally, Dw is the energy demand

that must be satisfied for demand scenario w.

The energy deficit Edefv is simply calculated by summing energy for the periods in

which energy were purchased, but where the EV has not yet plugged in:

Edefv =

T∑

t=1,avt=0

ctdt (7.7)

Here avt is the availability for the given time period t and plug in scenario v. Next,

the definition of a energy surplus becomes:

Esurvw = (

T∑

t=1

ctdt − Edefv +Kvw)−Dw (7.8)

86


Esurvw , the purchased energy minus the demand, is then the energy that can be sold

on the adjustment market.

Using the above equations the market costs become:

Anegwv = pA,negKvw (7.9)

Aposwv = pA,posEsur

vw (7.10)

Bposv = pB,posEdef

v (7.11)

In equation 7.9, 7.10 and 7.11 the market prices (pA,neg,pA,pos,pB,pos) are multiplied

with the energy deviations found at the time of plug in.

These cost calculations assumes that the fleet operator will be able to use the

adjustment market at the 2nd decision stage to compensate for most of the energy

deviations caused by the realized scenario for plug in and energy demand. An exception

is the positive deviation Edefw caused by energy being purchased in hours prior to plug

in. These will have to be handled through the balancing market and thus carries a

higher cost.

The above model have been implemented and tested in GAMS. If there is a large

probability of a late arrival and high demand, the fleet operator will tend to purchase

more energy in the day-ahead market and at a later point during the night, to avoid

adjustment and balancing market costs.

7.3.4 Introducing Uncertainty - Grid Constraints

Grid constraints were discussed in chapter three “Grid integration” and it was argued

that certain strategies could be used to provide the fleet operator with knowledge on

available capacity in the planning phase. This section describes how grid constraints

may influence the cost of charging.

In the following, grid constraints is defined as a reduction of the available active

power during one or more time periods due to congestion.

If the fleet operator has knowledge on grid constraints at the time of day-ahead

energy bids and purchases , then deviation costs can be avoided. The cost of charging

87

7. OPTIMIZATION

may, however, still increase if the constraints means that the fleet operator is unable

to utilize certain, low-price, periods.

If the grid constraints are only revealed during the operational phase, then the

fleet operator will most likely attempt to predict the periods where such constraints

may occur. These predictions can, as was done for the plug in uncertainty, be used to

generate a set of scenarios. The objective function defined in the previous section 7.5

becomes:

Min

W∑

w=1

πw

V∑

v=1

πv

O∑

o=1

πo

T∑

t=1

ctdtpt +Anegwv −Apos

wv +Bposvo +Bneg

o (7.12)

In 7.12 the summation∑O

o=1 is added and the balancing market costs have been

altered.

o is a scenario for grid congestion with probability πo. It can be seen that grid

constraints may incur additional costs in the balancing market for both positive and

negative deviations (Bposvo and Bneg

o ).

The following equation defines the energy deviation created by grid congestion:

F defvo =

T∑

t=1,avt=1,cot=0

ctdt (7.13)

The above sums up the contracted energy in hours where the EV is plugged in

(available avt = 1) but can not charge due to a charging constraint (constraint cot = 0).

The balancing market costs can now be expressed as:

Bposv = pB,pos(Edef

v + F defv o) (7.14)

Bnegv = pB,neg ∗ F def

vo (7.15)

The above equations, 7.14 and 7.15 , assumes that deviations caused by charging

constraints can not be handled through the adjustment market but must be handled

using the balancing market. The positive deviation caused by not being able to charge

due to grid constraints are added to the positive deviation created by late plug ins in

7.14. It is assumed that the energy missing from the battery, due to grid constraints,

88

7.4 Sub-conclusions

would still be needed and that the balancing market is used for such unplanned charging

7.15.

The above model have been simplified in that the costs of charging constraints only

applies to energy bought in the day ahead market, not the energy bough and sold in

the adjustment market.

The above grid constraint scenarios have been implemented in GAMS. High balanc-

ing costs will generally keep the fleet operator from purchasing energy in hours where

there is a risk of grid congestion.

7.4 Sub-conclusions




This chapter have described the use of optimization methods to minimize charging costs

for a fleet operator. It have been shown that:

• A binary optimization method can be used to minimize charging costs for EV

charging.

• Markets and stages can be introduced as a means for deviation recourse.

• EV owner-related uncertainty in terms of time of plug in and energy demand can

be managed using stochastic optimization with recourse.

• Scenarios for grid constraints can be included in the model.

The following have been done to implement and test the methods:

• In the EDISON fleet operator software, the binary optimisation method was

implemented via Microsoft Solver Foundation (MSF).

• The stochastic version of the optimisation problem have been implemented and

tested in GAMS.

89

7. OPTIMIZATION


The fleet operator will always have to manage charging based on individual and hard

constraints, and initial methods have therefore been formulated for a single vehicle.

There are, however, certain differences when considering an entire fleet. Energy devia-

tions can, for instance, be balanced out via running adjustments and rescheduling for

a larger group of EVs. Advanced portfolio management, and the real-time relationship

between individual and fleet level operations by a fleet operator, will be the subject of

further studies.

The stages presented here are rather course-grained and the fleet operator will most

likely include many more stages. Rescheduling and adjustments could be done every

15 minutes and/or at certain events. A scenario could be that if a specific vehicle have

not plugged in at 5 PM, the fleet operator would, according to its probability scenarios,

not expect the vehicle to plug in before 10 pm. Replanning could then be done at 5

PM rather than waiting until the plug in occurs as is done in the above model.

The model reflects that grid uncertainty would incur a higher cost than EV owner

uncertainty. This is because most EV owner-related uncertainty is resolved at the time

of plug in, while grid constraints may occur throughout the entire operational stage.

Also, all deviations caused by grid constrains may have to be settled via the balancing

market.

Future extensions to the proposed model could be the addition of more constraints,

e.g. battery, and uncertainties, e.g. price. Finally it may also be desirable to extend

the planning period to cover a longer period e.g. an entire week. This may better

incorporate variable patterns in EV owner behavior and market costs.

90

8

Conclusions

This chapter will summarize the results of each of the two parts of this thesis; Domain

interfaces and supporting informatics. Finally, it will be illustrated how the topics

investigated in the first part relate to the topics of the second part.

8.1 Domain Interfaces

The chapters, part of the domain interfaces investigation, have sought to uncover

the objectives and requirements of the three key stakeholders - market, distribution

grid, and Electric Vehicle (EV) owner. The chapters also demonstrate how such con-

siderations will impact the informatics solutions for intelligent EV integration. Results

for each stakeholder is covered in the following.

Market Integration

First, EV integration towards the existing spot market was investigated. Demonstra-

tion software was developed that showed that EVs could be charged in correspondence

with energy prices from the spot day-ahead market. Next, ancillary service markets

were investigated in a comparative study between ENTSO-E and American PJM mar-

91

8. CONCLUSIONS

kets. The outcome of this study was recommendations on EV Ancillary Service (A/S)

market integration and a set of proof-of-concept tests.

This study has shown that both the day-ahead energy markets and ancillary service

markets are possible targets for EV market integration.

It is recommended to pursue the market products (existing and new) that can best

match and reward the EV’s potential as a fast-response high power resource. Charging

in correspondence with dynamic energy prices may represent the low-hanging fruit and

the short-term aim of EV market integration. More advanced uses, including A/S

market participation, can both increase the complexity, but also the potential value.

Impacts on battery lifetime must be included in the equation for A/S provision.

Future research should focus on new markets and the best possible utilization of

EVs in the power system.

Grid Integration

The Thesis Author has sought to understand how grid congestion may influence the

operation of a fleet operator. A set of grid congestion mitigation strategies have been

suggested and evaluated in terms of complexity, risk and value. Operation maps with

functional stages have been employed as a tool to analyze the tasks and possible stake-

holder interactions that each strategy would entail.

The study has revealed the necessity of understanding and harmonizing the objec-

tives and operational tasks of stakeholders that would directly be impacted by con-

trolled EV charging. Early coordination of grid capacity sharing is found to benefit

both a fleet operator and the distribution system operator, and would better allow EV

participation in system-wide services.

It is recommend to consider whether the future distribution grid should be extended

to a degree where it can support charging schemes which would cluster charging in time.

This is a trade-off between the value of allowing EVs to target system-wide needs and

the value in reducing the need of local grid reinforcement.

Future research should continue to target grid congestion mitigation and the impact

congestion will have on the overall use of EVs in the power system.

92

8.1 Domain Interfaces

EV Owner

The Thesis Author has concentrated on plug in behavior and its impact on the flexibility

of controlled charging. First, a set of internal combustion engine vehicle data was used

to show how a prediction could be performed as input to an optimization method.

This approach was implemented and tested as part of the EDISON project. Next,

attention was given to a set of real EV driving data, and some early investigations of

plug in behavior, including the impact of seasonal changes and driver familiarization,

were conducted.

The study revealed a high degree of flexibility for the EVs and owners investigated;

The plug in periods were rather deterministic and stretched over long periods of time

compared to the energy demand they had to fulfil. The study has also hinted at some

of the external stimuli that this flexibility may be used towards. The results of the EV

owner plug in behavior investigation can be seen as an input to the optimization chapter

where historic data can be used to generate scenarios for stochastic programming.

The Thesis Author recommends using the analytic capabilities of a back end system

to predict and utilize the flexibility of an EV. This will strengthen the EV’s value as a

resource and can help reduce the required involvement of the EV owner as long as some

clear frame conditions for controlled charging are made. A benefit/risk trade-off must

be understood and accepted by the EV owner. The risk can be based on the empirical

distribution of plug outs and energy demand while the benefit depends on the target

of controlled charging.

Future research for the Thesis Author will include a more comprehensive investi-

gation on the Test-An-EV data. A better understanding of plug in behavior is key in

understanding the EV’s potential in the future power grid.

93

8. CONCLUSIONS

8.2 Supporting Informatics

The chapters part of the supporting informatics investigation are influenced by

what was found in the domain integration study. The findings are summarized in the

following:

Control Architecture

Based on studies of the virtual power plant concept, a centralized, commercial, EV

specific architecture was defined. Next, the functionalities that should be part of such a

platform were specified. Finally, the architecture and functionalities were implemented

in code and used in the demonstration of EV fleet management.

The architecture chosen is meant to reflect a viable and readily implementable

way of interfacing an EV fleet with the market. The architecture developed is largely

similar to a set of other large EV integration research projects investigated by the Thesis

Author. It is concluded that the suggested architecture is well suited to facilitate EV

integration.

It is recommended that original equipment manufacturers and fleet operators come

together in standardization to establish what systems and entities should be allowed to

influence charging. It can be recommended to focus on centralized control short-term

and then move towards a more decentralized scheme as research in this area matures.

Future research could focus on combinations of centralized and decentralized con-

trol and establish where controlling logic is best placed.

Communication

In this study IEC 61850 was investigated in regards to EV integration. It was sug-

gested to extend the IEC 61850 mapping with representational state transfer services

for increased simplicity and interoperability. For the data model a new set of logical

nodes was suggested to better accommodate EV and EV supply equipment hardware.

94

8.2 Supporting Informatics

These nodes have been sent to IEC for consideration. The IEC 61850 with REST

communication scheme have been implemented and tested as part of EDISON.

The study revealed that using the IEC 61850 standard is a valid option for monitor-

ing and controlling the charging of an EV. There are, however, a number of competing

standards which could challenge the use of IEC 61850.

The main recommendation is to use a standard that will not limit the use of the EV

in present or future scenarios. This means defining a broad and flexible information

model which will facilitate the necessary information and control. The standards also

have to meet certain requirements on security which can be ensured by using the right

protocols and following standardized security profiles.

Academia should follow standardization to ensure the continued support of future

smart grid concepts. It is also recommended to start the transition from the original

top-down SCADA approach to a service oriented architecture for EV integration.

Optimization

It was shown how optimization can be used as a tool to reduce the cost of charging.

First, a simple binary optimization method was proposed. Then the model was ex-

tended to include markets and the use of two-stage stochastic programming to include

uncertainty in regards to both EV owner plug in behavior and grid capacity.

The study showed how several constraints and considerations can be combined in

an optimization model and how uncertainty can be included in the form of scenarios

as part of stochastic programming. The model also illustrated that uncertainty in the

distribution grid and EV owner behavior can have an adverse effect on charging cost

minimization.

One of the main recommendations is to make optimization models that includes as

broad a set of stakeholder considerations as possible.

Future research by the author will extend the model with more constraints and

extend the use of stochastic programming. The continued examination of real EV data

can help shape and test future models.

95

8. CONCLUSIONS

8.3 Intelligent EV Integration - Closing Remarks

The following figure 8.1 illustrates the relationship between the chapter topics of this

thesis. More specifically it shows how findings and conclusions of initial chapters have

been picked up and reflected in later chapters.

Figure 8.1: Chapter topic relationships - Illustration of how information and choicesfrom the domain investigations map into the informatics chapters

The above figure, along with the findings described above, have shown how the two

objectives of this thesis have been fulfilled by conducting a thorough domain investiga-

tion and by identifying suitable informatics solutions.

It is hoped that the findings presented, both concerning stakeholders and solution

technologies, can be used to release the potential of EVs and promote and support the

intelligent integration of EVs into society.

96

9

Core publications

97

9. CORE PUBLICATIONS

A Coordination Strategies for Distribution Grid Conges-

tion Management in a Multi-Actor, Multi-Objective

Setting

This paper has been accepted for publication as part of the 3rd IEEE PES Innovative

Smart Grid Technologies, 2012, Berlin.

98

1

Coordination Strategies for Distribution GridCongestion Management in a Multi-Actor,

Multi-Objective SettingPeter Bach Andersen, Junjie Hu, Kai Heussen

Abstract—It is well understood that the electric vehicle asa distributed energy resource can provide valuable servicesto the power system. Such services, however, would have toco-exist with hard constraints imposed by EV user demandsand distribution grid operation constraints. This paper aimsto address the interactions between the stakeholders involved,mainly considering the distribution grid congestion problem, andconceptualize several approaches by which their diverse, poten-tially conflicting, objectives can be coordinated. A key aspect to beconsidered is the relationship between the operational planningand the handling of real-time events for reliable grid operation.This paper presents an analysis of key stakeholders in terms oftheir objectives and key operations. Three potential strategiesfor congestion management are presented and evaluated basedon their complexity of implementation, the value and benefitsthey can offer as well as possible drawbacks and risks.

Index Terms—Electric vehicle integration, Distribution grid,Congestion management, Smart charging

I. INTRODUCTION

Grid integration of electric vehicles, distributed generation,and other distributed resources has been a driver for a rangeof smart grid research activities. Here, the field of intelligentelectric vehicle (EV) integration is aimed at minimizing theadverse effects of introducing electric vehicles into the powersystem and maximizing the value for EV owners, the powersystem, and society as a whole.

A large part of intelligent EV integration research has beenaimed at such topics as optimal charging of electric vehiclesin term of charging cost [1]–[3], enabling renewable energy[4]–[6] as well as providing ancillary service to the powersystem [1], [7]–[9]. Such studies have primarily been aimedat system-wide power services and energy markets while notconsidering the distribution network. Concurrently, studieshave been carried out that look at charging management solelyfor the purpose of avoiding distribution level grid congestion[10]–[14].

Lately, research done in [15], [16] have been striving tocoordinate these objectives, i.e., to optimize the utilizationof electric vehicles while still respecting the hard constraintsimposed by consumers’ needs and distribution operation con-straints. In [15], a conceptual framework consisting of boththe technical grid operation and a market environment wasproposed to integrate EVs, the activities of all the actors

This work was supported in part by the iPower project.Peter, Junjie and Kai are with the Center for Electric Technology, Technical

University of Denmark, Denmark. Email: {pba,junhu,kh}@elektro.dtu.dk

including fleet operator (FO), distribution system operator(DSO) and consumers are described and the simulation resultsindicate that smart charging can maximize the EV penetrationwithout exceeding grid constraints. However, further researchon the coordination between FO and DSO and the interactionbetween FOs and consumers are not addressed clearly. Afurther development can be seen in [16], in which a complexscheduling problem involving consumer, fleet operatoor andDSO were analyzed. The results shows that both power andvoltage constraints due to electric vehicle charging can beavoided while the FO and consumer can achieve the objec-tive of minimizing charging costs and fulfilling the drivingrequirements. This approach requires a somewhat complexcoordination between DSO and FO but can potentially delivera very good solution in terms of optimal grid utilization andsafety.

This paper aims to add to the existing research by addressingthe interactions between the various actors and conceptualizeseveral approaches, by which their diverse, potentially con-flicting, objectives can be coordinated with respect to theoperational constraints of the low voltage distribution grid.A key aspect to be considered is the relationship between theoperational planning done by the actors and the handling ofreal time events which is vital for the DSO and the distributiongrid that it represents.

While this paper focuses specifically on the case of EVintegration, the coordination strategies presented, aiming atcongestion management in general, can to a large extendbe translated to a more generic demand side managementperspective.

The remainder of this paper is organized as follows: Sec-tions II presents three key stakeholders along with theirobjectives and operational tasks. In Section III a full mapof the operations identified is presented and Section IV thenexpends the map in the examination of three different coor-dination strategies. Finally, key contributions are summarizedand discussed in Section V.

II. ACTORS: OBJECTIVES AND OPERATIONAL TASKS

An overview of the actors, the grid and the main controloperations is presented in Figure 1. The figure conveys howthe actors’ operations are coupled through interactions via a)a common physical infrastructure, b) control relations andc) other information exchange. The coordination of theseoperations needs to reflect each actor’s objectives as well

2

Fig. 1. Actors (stakeholders), problem domain and main information andcontrol flows.

as operational constraints. In this problem formulation, wefocus on describing the following key stakeholders and theirobjectives.

• Distribution System Operator (DSO),• Fleet Operators (FO),• Customers (controllable loads / EVs).Other relevant influencers include the transmission system

operator (TSO), other market actors and conventional demand.Their influence is conveyed via control signals, market prices,and physical network utilization, respectively. They do nothave to be considered here explicitly as their role with respectto the distribution level is encapsulated via the DSO and FO.

In the following, these key actors are described in termsof objectives and the operations performed to satisfy theseobjectives.

A. Distribution System Operator

The main purpose of the distribution grid is to enablereliable power delivery to customers at a low-voltage level.Grid operation by the DSO is therefore aimed at effectivelybalancing two main objectives a) reliable grid operation and b)low cost of operation. We identify the following value driversfor a DSO:

1) Grid component investments,2) Capacity utilization factor,3) Component lifetime,4) Operation cost (incl. resistive losses),5) Instrumentation and automation efforts.Provisioning of distribution grid transfer capacity is planned

to be sufficient in all cases, that is, capacity is provided by thestandard of annual peaks, plus safety factors for anticipateddemand increases. In practice this means that distributiongrids tend to have a relatively low utilization factor. On theother hand, the distribution grid planners calculate with a high

‘diversity factor’: it could be safely expected that due to theindependent nature of most electricity consumption would leadto a smoothing effect that would reduce the absolute peak. Asa result, secondary transformers in the distribution grids canbe expected to be dimensioned at lower capacity than the totalcurrent capacity of all connected households.

The operating state of the distribution grid is limited by thefollowing operation constraints:

• Voltage limits (voltage quality),• Thermal limits of cables & transformers,• MVar bands (interface to TSO), or• Protection settings.

In this paper the focus is on the distribution grid’s ability totransfer active power.

1) Congestion management: The term ‘congestion’ in dis-tribution grids refers to a situation in which the demand foractive power transfer exceeds the transfer capability of thegrid. As the electricity grid cannot physically get congested,the term subsumes the complex mapping of the above men-tioned grid constraints to the network active power transfercapacity as seen for each connection point and the need fordeferring demand (or generation1). Whereas the constraintslisted above are specified in terms of limits for specificparameters (voltage, current, reactive power, active power),they all may influence the active power transfer capabilityavailable at a connection point. Their mapping is non-trivial,as it depends on properties of the physical infrastructure,characteristics of consumption devices and built-in controlbehaviours required by the respective grid code.

In general, the term ‘congestion mitigation’ can then beassociated with two types of strategies: a) to (locally) increasethe transfer capacity by means of reactive power and voltagecontrol and b) by coordinating the throughput via deferral orcurtailment of demand [17]. Both strategies aim at increasingthe utilization factor of the distribution grid.

Here, the term ‘congestion management’ explicitly refers tostrategies of type b), which aim at the coordination of activepower demand with respect to congested grid locations. Itcan be assumed that available strategies of type a) will beexhausted before type b) strategies are applied. Building onthe proposal in [17], the base case for congestion mitigationwill be considered active power curtailment.

2) Distribution System Operation Today: DSO tasks inconventional system operation, are mostly focused on ‘off-line’ tasks related to asset management and maintenance.Distribution systems today tend to be weakly monitored ascompared to transmission grids, and controlled in a decentral-ized fashion on the basis of preconfigured local controls (e.g.by means of grid codes and protection settings). Supervisorycontrol is then reduced time-of-day controlled adaptation ofcontrol settings, configuration management in response tooutages and maintenance related challenges.

Key Operations:• Grid dimensioning (incl. contigency planning and load

curve estimation),

1For the remainder of this paper, the perspective of distributed generationis implied.

3

• Maintenance and outage related topology reconfiguration,• Adjustment of transformer taps,• Fuses and relay operation,• Fault-analysis and repair.3) Operations in active distribution grids: To illustrate a

future operation scenario with a higher level of automation,it is considered how the above operations can be extendedwith additional online- and data intensive operations. In orderto identify and solve congestion problems, the DSO requiresadditional measurement equipment and/or technology enablingthe anticipation of load patterns and grid ‘bottlenecks’.

Key Operations for DSO congestion management in ‘ad-vanced’ distribution grids:

• Demand forecasting• Grid state estimation• Online grid measurements• Real-time intervention in case of unexpected deviations

challenging grid reliability• Meter data collection and aggregation

economically and reliably and shows a relation between VPPsand DSO. In this control system, several families are suppliedunder one feeder and they own controllable devices, i.e,electric vehicles, besides some conventional load, such as light,TV etc. For these controllable devices, they are divided intotwo groups according to the method controlled by the VPP,one group is directly controlled by the VPP, which meansan extra cards or relays are installed on the user’s device,and the VPP can turn on/off the devices; another group iscontrolled by price, in which the devices are assumed to beprice-responsive. VPP starts to make an energy schedule forits customers with the purpose of minimizing the electricitycost and meanwhile fulfilling their requirement. This problemcan be formulated as a linear programming or dynamicalprogramming way [1], [2]. The congestion problem mayfirst happened during the scheduling making, this problemshould be solved by the coordination between DSO and VPPs.After the charging schedule was set up, ideally, the users areexpected to totally follow the schedule. However, in general,deviation may happen. With the purpose of avoiding thepossible congestion (happened again in real time), DSO willmonitor the system’s operation conditions dynamically andcoordinate with VPPs. The following subsection will discussthe mechanism of solving these congestion problems.

B. Fleet Operators

The fleet operator (FO) is a commercial entity that aggre-gates a group of EVs in order to actively integrate them intothe power market, and in so doing, utilizing their chargingflexibility to meet a financial goal. The financial goal couldbe to achieve savings on the purchase of energy or makeearnings by selling ancillary service products or, possibly, acombination of the two.

A FO follows the concept of a ’virtual Power Plant’which was first introduced to allow market participation fordistributed energy resources.

In the current European power and energy markets, the FOcould be a retailer with either a load balance or production

balance responsibility, depending on the market/service thatthe FO would address.

The value drivers for a FO are:1) Maximize profits or minimize costs by participating in

markets.2) Providing services (cost reductions, convenience etc.) that

will attract EV owners as customers.Due to the participation in markets and customer services,

the FO is subject to operating constraints defined by contrac-tual commitments:

• Market schedule (energy/h)• Customer demand (driving needs)• TSO driven ancillary service requirements (e.g. reserve

capacity)How the economic value obtained through the market is

shared with the customers would be business case specificto the FO. It is also assumed that the FO would maintain aService Level Agreement (SLA) with its customers that woulddictate the degree to which it may control and manipulate theEV charging patterns to achieve goals other than customerdriving. This would represent a trade-off between energysavings and EV driving availability that should be understoodand accepted by the customer.

The operations of the fleet operator can be divided into fleetlevel operations and individual level operations as follows.Fleet level:

• Selection of market products and services• Contracting• Market/service forecasting

Individual level:• Customer SLA management• Driving pattern prediction

C. Customers

The customers, here EV owners and drivers, are not as-sumed to be particularly interested in grid issues. Their mainvalue drivers are expected to be:

1) Availability of EV for driving2) Total cost of ownership/energyIt is assumed here, that the customer will opt for conve-

nience and delegate most of the charging control to the FO.The customer is expected to rely on the frame conditionsexpressed in the SLA for the daily charging managementfor ’typical’ and predictable driving patterns. An optionalfeature would be to let the customer communicate his or hersexact driving intentions to the FO. This would strengthen theFO ability to utilize the specific EV’s flexibility. The mainoperations of the customer, besides transportation, would thenbe:

• Accept, and possibly modify, the SLA with the FO.• Inform the FO of any non-typical driving needs.

III. MAP OF OPERATIONS

The operations outlined above will in this section be mappedgraphically to enable an analysis of different coordinationstrategies.

4

Fig. 2. Map of base-case operations.

A. Analysis FrameworkIn this section a classification approach is introduced, based

on the understanding that the coordination approach is both anautomation problem and a market design problem.

A widely accepted hierarchical decomposition of processcontrol into four functional levels describes integrated indus-trial automation [18]:

• Level 4: plant(s) management• Level 3: production scheduling and control• Level 2: plant supervisory control• Level 1: direct process control

This level-hierarchy is associated with several characterizingparameters, including e.g. time scale, time resolution, planninghorizon, and hierarchical dependency of objectives. No singleone of these parameters can be considered directly decisivefor forming the levels, but together they generate the needfor distinguishing qualitatively different levels of automation.The hierarchical dependency of objectives, i.e. that one level ishigher and another is lower in ordering, is associated with themeans-ends structure of objectives: A higher-level objective isbroader in scope and more closely associated with the businessobjectives of the respective process, and thus ‘higher’ in thevalue chain; a lower-level objective, in contrast, is there tosupport and enable other process functions.

In the present multi-actor, multi-objective setting, the sin-gle hierarchy does not hold: different actors have differentobjectives, and yet they must interact with respect to the sameprocess. The present means-ends perspective shall be strippedfrom the automation hierarchy, to allow for a high-level mapof operations. Key elements to be captured in the new mapare:

• Key operations and their allocation to actors• The association of operations with a time scale including

a distinction of operational and administrative functions

• The possibility to map interaction sequences betweenoperations

Removing the means-ends perspective, we are left with amostly time-driven decomposition. We consider the followingfundamental stages:

I. Offline PlanningII. Online Scheduling

III. Real-time Operation (Execution)IV. Offline Settlement

These stages model a logical sequence: each stage is basedon a completion of the previous stage. The timing aspect is notessential here as certain types of operation can be performedfaster with improved technology. The stages are characterizedin the following:

Settlement is about the aftermath: recordings (measure-ments, sent commands, etc.) of executed operations are con-solidated and (financial) responsibility is allocated. The op-eration stage is about pure execution in real-time. Plans areonly executed, and unplanned events occur and physical aswell as automatic controls respond without deliberation. The’online’ scheduling stage can in time be closely coupled withoperation (e.g. reactive scheduling with a 5min resolution) orextend hours or days ahead of it. Scheduling is the stage inwhich available resources are best known and the platformfor execution is to be prepared. Finally, the first stage, herecalled ’planning’ has been distinguished from scheduling inthe same fashion as unit commitment is distinguished fromdispatch: Depending on the specific coordination strategy, wedistinguish operations that can be coordinated in a ad-hocfashion and those that provide the basis for such ad-hoc deci-sions. Due to these clear distinctions, the framework supportsthe discussion of interactions between key operation tasks forcross-stakeholder coordination for the complete process. Asthe operations can be associated with operation objectives of

5

the respective stakeholder, this map allows for an analysis ofthe incorporation of the respective value drivers by a givencoordination strategy. This ’horizontal’ level means that theoperations have to be considered at the same level of abstrac-tion. A ’vertical’ perspective would unfold more details of theoperations, eventually also revealing physical interactions [19].The goal is to analyze the benefits and trade-offs involved inspecific coordination strategies. Value is hereby understood ina generic sense as to contribution to a stakeholder’s objectives.Given the Operation-Stakeholder allocation and the analysisof value drivers, a similar framework can be employed to alsoanalyse value-network constellations, as exemplified in [20].However, that type of analysis is beyond the scope of thispaper.

B. Base Case Map

To establish a firm foundation for the analysis of differentmarket-based coordination strategies, we identify a base casewith a minimum set of operations that will be common toall considered congestion-coordination strategies (Figure 2).This base-case maps out the operations DSO, FO and EVowner would be required to execute in either of the coordinatedcongestion management schemes.

The base case uses the following assumptions:- As discussed in [15] and [19], the introduction of con-

trollable demand with significant power capacity such as thatof electric vehicles implies a significant risk for distributionassets. To avoid potentially harmful charging configurations,we include the concept of an ’emergency brake’ in all EVcharging post: it enables the unconditional interruption of EVcharging on request by the DSO. It could be implemented onthe basis of a ’keep-alive’ signal, the failure of which wouldimmediately interrupt the EV charging process

- The maps allocate all optimization and coordination intel-ligence to the FO. It is understood that many of the operationscould be implemented using distributed algorithms e.g. in theelectric car or charging post.

- Vehicle-to-Grid (V2G) is not considered in this publica-tion. The technology’s potential for congestion relief and itsimpact on power quality are, however, relevant for conges-tion management and should be further addressed in futurepublications.

IV. COORDINATION STRATEGIES FOR CONGESTIONMANAGEMENT

Approaches to the congestion problem are outlined and thenclassified and analysed using the map described in the previoussection. All three strategies represent very new approaches todistribution grid congestion management and none of themhave been investigated in very great detail.

The strategies investigated are:• Distribution grid capacity market• Advance capacity allocation• Dynamic grid tariffFor each approach a new map is drawn where operations

required specifically for the strategy in question are presented

Fig. 3. Distribution Grid Capacity Market.

in bold. Shared supporting operations beyond the main traceof operations have been omitted for compactness.

To describe how technically and administratively demandingit would be for a DSO or FO to implement and operate the newprocedures required by the coordination strategy the parametercomplexity is used. A second parameter value denotes the de-gree to which the strategy would help the stakeholder achieveits operational goal. Finally, the parameter risk describespotential problems associated with the respective strategy incontext of a currently uncertain external environment.

A. Distribution grid capacity market

As proposed in [21], this strategy would require a newmarket for trading distribution grid capacity. For this paper theterm ‘Distribution Grid Capacity Market’ is used; Also a newrole ’market operator’ is introduced which is responsible formarket operation. The FO will submit requests for their ’aggre-gated schedule’ consisting of their scheduled consumption foreach node (aggregated capacity), in response they will receivea price for each node, reflecting the respective congestion, andare requested to update their charging schedules. The processis iterated until all constraints are satisfies. The concept usedin this strategy can be found in a similar form for the powertransmission system [22].

1) Operation sequence:• First, the FO will make an aggregated energy schedule

for EV owners based on its objectives. Afterwards, thisaggregated schedule will be sent to the market operator.

• The Market operator will generate a price for the grid ca-pacity according to the schedules. This price is associatedwith the power difference between the sum of scheduledpower and upper power limits of the grid.

• FOs would inform the market operator of their newenergy schedule under the initial price. The schedule canbe calculated based on the marginal value of a utilityfunction, e.g., cost function in term of the power deviationor satisfaction degree with the ’preference difference’(the difference between energy schedule after congestionmanagement and energy schedule before congestion man-agement).

• The market operator then determines whether the distri-bution grid is overloaded or underutilized and comes upwith a new corresponding price. After a certain number

6

Fig. 4. Advance Capacity Allocation.

of iterations, the price will eventually converge andaccepted by all FOs, which establishes a binding chargingschedule.

2) Evaluation: Complexity:With this strategy, a new market is required which means

that the corresponding platform for trading grid capacity needsto be designed and implemented. Also, new communicationflows are needed to support the market operations. The marketitself will be rather complex to establish and operate.

A lot of complexity is transfered from DSO to the capacitymarket. Here the DSO will be required to provide the measuredand estimated power information to the market operator. TheFOs will take on the task of trading capacity and reschedulingthe energy consumptions for their customers etc..

Value: With this new market, FOs will have more flexibilityto trade and utilize the grid capacity of a distribution system.If the market and capacity information is reliable and well-designd it will ease the operation of the DSO, enable acomparatively high utilization factor and reliable schedules forFOs as well. A further benefit is that no actual consumptioninformation is revealed to other market parties, as only acommon congestion price is established per node.

Risk: It must be guaranteed that all FOs adhere to the rulesof the market. Another risk lies in the algorithms used toarrive at prices based on utility functions i.e. the computationalrequirements and time needed for a solution to converge.

B. Advance Capacity Allocation

The simple concept behind this strategy is that the DSOcould identify and pre-allocate available capacity by defininga conservative static capacity limit (kW) for each feeder-linebased on the capacity rating of the respective transformers andcables and the expected conventional load curves. The EV-equipped households attached to a certain feeder-line wouldthen be given a certain share of available capacity which wouldbe allocated to the FO representing them. To avoid inefficientutilization of available grid capacity due to unused capacityshares, a second step is added to the strategy where FOs cantrade their allocated capacity in an over-the-counter manner.

1) Operation sequence:

• The ’Contracting for capacity sharing’ operation wouldinvolve letting the DSO know the mapping between grid

connected EV-equipped customers and FOs and thendetermining how capacity is shared.

• During the scheduling stage, the DSO would via grid loadforecasting estimate the available capacity and communi-cate this to the FOs as defined in the contracts establishedin the planning stage. After having received its share, theFO could then optionally engage in capacity trading withother FOs operating on the same feeder.

• The DSO should be informed of the bilateral capacitytrading so that, in case of violations (i.e. total load ob-served from EV charging in specific part of grid exceedssum of allocated shares) penalties for violations can beappropriately placed at the responsible FO.

• Finally the strategy would involve settlement both be-tween DSO and the individual FO and possibly aninternal settlement between the FOs that engaged in thebilateral trade.

2) Evaluation: Complexity: Here, rather than dimension-ing the physical characteristics of the grid depending on loadprofiles and simultaneity factors, the DSO would limit thecontrollable load based on the physical characteristics of thegrid. In addition to the location-based grid capacity the DSOwould also need to map each grid customers endpoint to anassociated FO. There is also some complexity in how theFOs will trade capacity internally and how violations of gridcapacity will be dealt with in the settlement stage when tradinghas been involved.

Value: This strategy represents a rather simple coordinationmechanism between FOs and DSOs. The DSO is only requiredto communicate a single value (capacity) to each FO and isthen removed from the equation until the settlement stage. Thiswill simplify the responsibilities of the DSO considerably andleave the detailed capacity allocation to the entities directly incontrol of EV charging i.e. the FOs. There are also advantagesto the FO since it will see a guaranteed capacity, free fromstochasticity, early in the planning stage. Early information isvaluable to an FO attempting to optimize charging to meet avariety of goals such as market services and individual drivingneeds.

Risk: There is the risk that a single kW limit set-point pergrid node is too crude a mechanism to handle thermal loading- any unexpected change in base load during operation mayvoid the DSO’s estimation of capacity shares which has beenhanded out to the FOs during scheduling. The risk in thisapproach also lies in the effectiveness and reliability of theFO bilateral capacity trading. If the FOs can not be trustedto handle the management and trading of capacity amongthemselves, there will be the need of a more formal framework,e.g a market, and new definitions of responsibilities, such asthe balance responsible parties seen in the energy market.

C. Dynamic Grid Tariff

In this solution, the distribution system operator generatesa time and grid-location dependent price for grid usage basedon expected nodal consumption levels.

The DSO anticipates the size and the price-responsivenessof the load at critical grid nodes and calculates the price to

7

Fig. 5. Dynamic Grid Tariff.

optimally reflect the expected congestion problem. FOs willthen see a dynamic nodal tariff and can make an optimalschedule with respect to the e.g. spot price and dynamic gridtariff.

The method considered here has been presented in [23].1) Operation sequence: The key operation aspects for this

coordination strategy are outlined in Figure 5.• In the planning stage, a distribution system operator

would create models for the price-sensitivity of relevantdemand clusters. These models would be updated ona regular basis based on learning from smart meterfeedback.

• In the scheduling stage the forecasted demand, gridsituation and present spot market prices will be employedto calculate appropriate branch prices for distribution gridutilization.

• The dynamic tariff is published to subscribers. Theadapted branch prices are received by the fleet operatorand employed to compute an optimal charging plan.

• During the operation stage, the charging schedule is ex-ecuted. In case of severe underestimation or fluctuationsof the actual demand, DSO controlled interruptions mayoccur in real-time.

• For settlement the timed consumption data is collected bythe responsible DSO and the published prices will thenbe employed to bill the actual grid usage individually.

2) Evaluation: Complexity: The main characteristic fea-ture of this approach is the simplicity of the interactions andalso the simplicity of integrating simple prices in distributiongrids.

The implementation complexity is high on the side ofthe DSO. This scheme cannot be established safely withoutinterruptability of the vehicle charging.

For the Fleet Operator–Consumer interaction, the establish-ment of a satisfactory service quality may require a specialattention to potential bottlenecks in the system from the sideof the Fleet Operator.

Value: As compared to the base case, this model enablesan increase of the grid utilization factor. The small number ofparticipants at a feeder level means that random behaviour(fluctuation consumption level) might be stronger than theprice sensitivity of the controllable demand. Even though theincrease of the utilization factor is therefore highly uncertain,the simplicity of the approach could justify its implementation.

TABLE ISTRATEGY OVERVIEW

Actor Complexity Value RiskDistribution Grid Capacity Market

FO High High LowDSO High High Low

Advance Capacity AllocationFO Medium High LowDSO Low Medium Medium

Dynamic Grid TariffFO Low Low HighDSO Medium Medium High

For fleet operators and consumers, the benefits are alsoindirectly associated with the increased grid utilization. Afurther benefit can be seen in the flexibility this approach offerswith respect to integrating other flexible demand units, as theprice, in theory, could interpreted by any unit.

Risk: It is unclear whether a meaningful price-sensitivityof demand can be established.

There is a risk that there is no ‘right’ price to avoidoverloading, if a sufficient number of EVs is connected tothe same feeder, there is no way for them to negotiatecapacity utilization in the given framework. Due to the re-quired interruptability, the high chance for unplanned charginginterruptions also implies an additional risk is on the side ofthe Fleet Operator / EV Owner.

V. DISCUSSION AND CONCLUSION

This paper has investigated the concept of congestionmanagement for distribution grids, detailing the operationsand interactions of two main stakeholders in three differentcoordination strategies. The purpose of the analysis was tohighlight the cross-actor dependencies that each such strategyentails along the operation timeline, and thus to globally assesscomplexity, value and risk for each strategy.

Table I summarizes these evaluation parameters across thestrategies and stakeholders. In this table the customer isrepresented by the FO.

The ‘distribution grid capacity market’ is expected to offerhigh value and low risk for both FO and DSO assuming aformalized, optimal and secure framework supplied by a well-designed market. such a market, however, may represent themost complex strategy to implement, which may hinder ordelay its real-life implementation.

‘Advance grid capacity’ is relatively easy to implement,but would require over-the-counter trading to efficiently useavailable grid capacity. The strategy removes complexity fromthe DSO but some risk may have to be managed due to thebilateral FO trading and the advance capacity allocation mightrequire more conservative estimates of the available capacity.The FO gains high value from early information on capacityavailability.

‘Dynamic tariffs’ would also be easier to implement thana capacity market, but may prove challenging to the fleetoperator due to added uncertainty and a possible conflict withsystem-wide smart charging schemes. It could also imposesome extra risk for the DSO to rely on prices rather than hardcapacity limits when considering individual feeder lines.

8

A few general observations can be summarized as:• Grid considerations will have to co-exist with other

objectives in the charging management of EVs.• Coordination and information exchange in earlier stages

can reduce complexity and benefit both the FO and DSO.• There may be a trade-off between ease of strategy

implementation and optimality. A compromise may benecessary for the first real-life implementations.

• A suitable strategy for coordination between DSOs andFOs will improve each stakeholders ability to reach itsobjectives considerably.

The analysis framework developed in this paper can beconsidered sufficiently generic for analysing operations withrespect to other distributed resources.

An important mechanism included in this paper is the ‘real-time intervention’ functionality used by the DSO. This last-resort ability to directly and immediately reduce or disconnectcharging may be a prerequisite for the deployment of effectivecoordination strategies.

In the end, it is hoped that this paper contributes to a betterunderstanding of the multifaceted challenge of EV chargingand helps the development of open, robust, and meaningfulstrategies for low voltage grid congestion management.

VI. REFERENCES

[1] N. Rotering and M. Ilic, “Optimal charge control of plug-in hybridelectric vehicles in deregulated electricity markets,” IEEE Transactionson Power Sy, 2010.

[2] O. Sundstrom and C. Binding, “Optimization methods to plan thecharging of electric vehicle fleets,” in Proceedings of the InternationalConference on Control, Communication and Power Engineering, 2010,pp. 28–29.

[3] T. Kristoffersen, K. Capion, and P. Meibom, “Optimal charging ofelectric drive vehicles in a market environment,” Applied Energy, 2010.

[4] F. Tuffner and M. Kintner-Meyer, “Using electric vehicles to meetbalancing requirements associated with wind power,” Pacific NorthwestNational Laboratory (PNNL), Richland, WA (US), Tech. Rep., 2011.

[5] J. Pecas Lopes, P. Rocha Almeida, and F. Soares, “Using vehicle-to-gridto maximize the integration of intermittent renewable energy resourcesin islanded electric grids,” in Clean Electrical Power, 2009 InternationalConference on. IEEE, 2009, pp. 290–295.

[6] J. Lopes, F. Soares, P. Almeida, and M. da Silva, “Smart charging strate-gies for electric vehicles: Enhancing grid performance and maximizingthe use of variable renewable energy resources,” in EVS24-The 24thInternational Battery, Hybrid and Fuel Cell Electric Vehicle Symposium& Exhibition, Stavanger, Norway, 2009.

[7] W. Kempton and J. Tomic, “Vehicle-to-grid power implementation: Fromstabilizing the grid to supporting large-scale renewable energy,” Journalof Power Sources, vol. 144, no. 1, pp. 280–294, 2005.

[8] ——, “Vehicle-to-grid power fundamentals: Calculating capacity and netrevenue,” Journal of Power Sources, vol. 144, no. 1, pp. 268–279, 2005.

[9] S. Han, S. Han, and K. Sezaki, “Development of an optimal vehicle-to-grid aggregator for frequency regulation,” IEEE Transactions on SmartGrid, vol. 1, no. 1, pp. 65–72, 2010.

[10] G. Heydt, “The impact of electric vehicle deployment on load manage-ment straregies,” Power Apparatus and Systems, IEEE Transactions on,no. 5, pp. 1253–1259, 1983.

[11] ——, “The impact of electric vehicle deployment on load managementstraregies,” IEEE Transactions on Power Apparatus and Systems, no. 5,pp. 1253–1259, 2007.

[12] R. Green II, L. Wang, and M. Alam, “The impact of plug-in hybridelectric vehicles on distribution networks: A review and outlook,”Renewable and Sustainable Energy Reviews, vol. 15, no. 1, pp. 544–553, 2011.

[13] J. Lopes, F. Soares, and P. Almeida, “Identifying management pro-cedures to deal with connection of electric vehicles in the grid,” inPowerTech, 2009 IEEE Bucharest. IEEE, 2009, pp. 1–8.

[14] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of chargingplug-in hybrid electric vehicles on a residential distribution grid,” PowerSystems, IEEE Transactions on, vol. 25, no. 1, pp. 371–380, 2010.

[15] J. Lopes, F. Soares, and P. Almeida, “Integration of electric vehicles inthe electric power system,” Proceedings of the IEEE, vol. 99, no. 1, pp.168–183, 2011.

[16] O. Sundstrom and C. Binding, “Flexible charging optimization forelectric vehicles considering distribution grid constraints,” Smart Grid,IEEE Transactions on, vol. 3, no. 1, pp. 26–37, 2012.

[17] T. Sansawatt, L. Ochoa, and G. Harrison, “Integrating distributed gen-eration using decentralised voltage regulation,” in Power and EnergySociety General Meeting, 2010 IEEE, july 2010, pp. 1 –6.

[18] D. Popovic and V. Bhatkar, Distributed computer control for industrialautomation. CRC, 1990.

[19] K. Heussen, “Control architecture modeling for future power systems,”Ph.D. dissertation, Technical University of Denmark, 2011.

[20] K. Kok, Z. Derzsi, J. Gordijn, M. Hommelberg, C. Warmer, R. Kam-phuis, and H. Akkermans, “Agent-based electricity balancing withdistributed energy resources: A multiperspective case study,” in Pro-ceedings of the 41st Annual Hawaii International Conference on SystemSciences, 2008.

[21] B. Biegel, P. Andersen, J. Stoustrup, and J. Bendtsen, “Congestionmanagement in a smart grid via shadow prices,” in Proceedings of the 8thIFAC Symposium on Power Plant and Power System Control, Toulouse,France, Sep. 2012.

[22] S. Stoft, Power system economics. IEEE press Piscataway, NJ, 2002.[23] N. O’ Connel, Q. Wu, J. stergaard, A. Nielsen, S. Cha, and Y. Ding,

“Electric vehicle (ev) charging management with dynamic distributionsystem tariff,” in proceeding of 2011 ISGT, 2011.

VII. BIOGRAPHIES

Peter Bach Andersen has a M.Sc. in Informaticsand is currently finalizing a PhD at the Departmentof Electrical Engineering within the Technical Uni-versity of Denmark (DTU). His research focus is onsmart grids in general but with a special emphasis onelectric vehicle integration. Peter is an IEEE studentmember.

Junjie Hu received the Master of Engineeringin Control Theory and Control Engineering fromTongJi University, ShangHai, China, in 2010. Cur-rently, he is a PhD student at the Department of Elec-trical Engineering within the Technical University ofDenmark (DTU).His research focuses on integratingcontrol policies on controllable load, mainly electricvehicle, for active power distribution system, ofwhich the control policies are direct control andindirect (price) control. Junjie is an IEEE studentmember.

Kai Heussen is assistant professor at the TechnicalUniversity of Denmark, where he also obtained hisPhD on Control Architecture Modeling for FuturePower Systems. He received his Dipl.Ing. in En-gineering Cybernetics in 2007 from University ofStuttgart. His current research focus is the design ofheterarchical and service-oriented control architec-tures for distributed control of power systems, withspecial attention to functional modeling and decisionsupport for automation design.

B A Comparison of Electric Vehicle Integration Projects

B A Comparison of Electric Vehicle Integration Projects

This paper has been accepted for publication as part of the 3rd IEEE PES Innovative

Smart Grid Technologies, 2012, Berlin.

107

1

Abstract -- It is widely agreed that an intelligent integration of electric vehicles can yield benefits for electric vehicle owner, power grid, and the society as a whole. Numerous electric vehicle utilization concepts have been investigated ranging from the simple e.g. delayed charging to the more advanced e.g. utilization of electric vehicles for ancillary services. To arrive at standardized solutions, it is helpful to analyze the market integration and utilization concepts, architectures and technologies used in a set of state-of-the art electric vehicle demonstration projects. The goal of this paper is to highlight different approaches to electric vehicle integration in three such projects and describe the underlying technical components which should be harmonized to support interoperability and a broad set of utilization concepts. The projects investigated are the American University of Delaware’s V2G research, the German e-mobility Berlin project and the Danish EDISON project.

Index Terms – Electric Vehicles, Distributed Energy Resources, Smart Grid, Standardization, Information and Communications Technology.

I. INTRODUCTION ommon to most electric vehicle (EV) power system integration projects are that they aim to prove that the EV,

when seen as a distributed energy recourse, can offer value and services that exceeds its primary function as a means of transportation. The value and services investigated can, for instance, be the minimization of charging costs, adherence to grid constraints or adjustment of charging behaviour to renewable energy production [1]-[3]. In practise, however, projects often differ in how the above is actually achieved.

A standardized approach to electric vehicle integration has many benefits including the support of roaming and a consistent support of advanced grid related services.

A good input for standardization is the research and demonstrations already done in the area of EV integration. For three such projects we use a set of parameters to categorize and compare the approaches and components used in each.

The parameters used in this paper are:

• Market integration and EV utilization concept • Architecture • Technology and standards

The work is supported by the project “Electric vehicles in a Distributed

and Integrated market using Sustainable energy and Open Networks', EDISON. (ForskEL Project Number 081216). Mr. Andersen and Mr. Garcia-Valle are with the Centre for Electric Technology, Department of Electrical Engineering, Technical University of Denmark, 2800 Kgs. Lyngby, Denmark (e-mail: [email protected]). Mr. Kempton is with the University of Delaware, College of Earth, Ocean, and Environment, USA.

By 'market integration' is meant the way that the EV, directly or indirectly, is connected to the power market to generate savings and possibly revenue for the EV owner. The number and composition of market stakeholders involved depends heavily on the business models and market environments under consideration. Market integration approaches could both be aimed at current energy and power markets or future, more dynamic, near real-time markets using price signals. The latter is researched in the ECOGRID.EU project [4].

Related to market integration is the 'utilization concept' in question. 'Utilization concept' covers which functionalities the EV will support and can roughly be categorized into the two following categories:

• Smart charging - Where the charging of an EV battery is delayed or advanced in time based on energy costs, grid constraints or renewable contents. Essentially the EV is used as a controllable load.

• Energy backup and power services - The Vehicle-to-Grid (V2G) concept, where power is delivered back to the grid from the EV battery, belongs in this category. V2G can increase the EV’s ability to provide regulating services.

The market integration and utilization concept chosen shapes the architecture of a project. By 'architecture' is meant the stakeholders and mechanisms used to influence the EV’s behaviour and interface it with the power system and market.

Three different 'generic' architectures are proposed by You Shi et. al. [5] for the concept of Virtual Power Plants, which also represents the architectural options in EV integration.

• Centralized control - In which a single entity (an aggregator) directly controls the behaviour of a group of electric vehicles.

• Partly centralized - In which a negotiation is taking place between the individual EV and the aggregator. The aggregator would thus only indirectly influence the behaviour of the EV.

• Fully distributed - In which the EV behaves as an autonomous and intelligent agent that could interface directly with the power grid and/or the market.

The fully distributed architecture is strongly related to the market integration concept of price signals while projects aimed at existing markets and grid dispatch systems tend to go for the more centralized architectures.

The final parameter used for the comparison is ‘technologies

A Comparison of Electric Vehicle Integration Projects

Peter Bach Andersen, Rodrigo Garcia-Valle, Willett Kempton

C

2

and standards’. The technology descriptions of the projects cover two topics. First, the components in both soft- and hardware which has been developed to support the computation and logic necessary for managing smart or bi-directional charging are mentioned. Secondly, the communication protocols used for transferring data between the entities are listed.

Related to technologies are the standards used by the projects. The standards can be subdivided into those dealing with communication and information models and those that deal with equipment and hardware.

The remainder of the paper is organised as follows. The following, 'EV integration projects' section will describe the EV integration approach used in three different EV integration projects by using the parameters defined above. Next, the 'Project comparisons and recommendations' section will highlight the similarities and difference between them and then explore the potential of standardization. Finally, the paper will summarize its findings and suggest future focus areas in the 'Conclusion' section.

II. EV INTEGRATION PROJECTS This section will introduce two European and one American

EV integration project. They have been chosen since they represent some of the biggest and most innovative research projects within the field. These are however far from the only ones and projects such as MERGE [6] (Mobile Energy in grids of electricity) could also have been included. Such projects represent mature investigations where a lot of experience have already been accumulated. The experience gathered is, apart for an input to standardization, also a good point of departure for a new generation of EV projects. Among the absolutely largest among the newest generation of projects are the European ‘Green eMotion’ project [7] with more than 40 partners and a 42 Mio. EUR budget. In the states the huge ‘The EV Project’ [8] represents the largest EV charging infrastructure project to date counting more than 60 participants and including thousands of EVs.

The three projects specifically chosen in this paper all share certain traits when it comes to the integration approach followed. They are all 'economic' integrations where money is earned through market participation. They also share a focus on existing markets and implement an either centralized or partly centralized architecture. Such similarities could be seen as a prerequisite for arriving at common solutions applicable for all such projects. There are, however, still many differences in the implementations which illustrates the challenges to standardization.

This paper focuses on four main entities which are present in all three project; The EV, the EV User, the Electric Vehicle Supply Equipment (EVSE) and the more generic Aggregator role, which would represent the 'interface' between a group of EVs and the power system or energy market.

As will be shown by the following, the projects differ in by which means the above entities should communicate, what information should flow between them and, in the end, which entity will control the behaviour of the EV.

A. The Edison project EDISON [9]-[10] is short for 'Electric vehicles in a

Distributed and Integrated market using Sustainable energy and Open Networks' and is a research project partly funded through the Danish Transmission System Operator (TSO) - Energinet.dk.

The goal is to develop optimal solutions for EV integration, including network issues, market solutions, and optimal interaction between different energy technologies. The technical platforms developed by EDISON should be globally applicable and is tested on the Danish island of Bornholm.

The EDISON consortium consists of the Danish utilities DONG Energy and Østkraft, the Danish Technical University (DTU), as well as IBM, Siemens, Eurisco and the Danish Energy Association. The three year project will conclude in 2012 but might be followed by an EDISON 2 project. The project webpage can be found at www.edison-net.dk.

Fig. 1. The architecture of EDISON

1) Market integration and EV utilization concept While several market integration concepts are within the research scope of EDISON, the initial focus is on the current Nordic NordPool market. Within NordPool the EVs can either be connected to the day-ahead energy market, be used for intraday regulation or for reserves. The first project phase puts its emphasis on the first and indirectly connect the EVs with the day-ahead spot market by controlling the charging in correspondence with hourly energy prices. This follows the smart charging utilization concept. 2) Architecture

The setup shown in figure 1 represents the implementation done in EDISON. The setup uses a centralized architecture where an aggregator, called the 'fleet operator' in EDISON, directly controls the charging patterns of the EV to facilitate smart charging. The conceptual role of a fleet operator could be maintained by any commercial party willing to adhere to the requirements of the Nordic power market. An EV in EDISON is seen as relatively simple with little local

3

intelligence. The argument is that most OEM EVs initially will lack the capabilities for local optimization. The EV needs only to implement an interface that would allow the fleet operator to extract status information, such as state-of-charge, and possible constrains set by the OEM. In Edison the charging spot would play the role of a 'proxy' in that it would extract EV information and manage smart charging on behalf of the fleet operator. The user will in EDISON communicate her or his charging preferences directly to the aggregator. 3) Technologies and standards

Between EV and EVSE, EDISON utilizes the IEC 61851 standard which describes the charging of EVs using different AC or DC power voltages over a conductor using on- or off-board equipment. Apart from specifications for equipment interoperability and safety, the standard also defines simple EV-EVSE communication via a control pilot wire using a pulse width modulated (PWM) signal with a variable voltage level. This allows the EV to communicate its state to the EVSE. IEC 61851 is used in EDISON since it helps satisfy safety requirements and will improve interoperability.

The IEC 62196-2 Type 2 Mennekes plugs are used on the conductor connecting EV and EVSE.

Apart from the control pilot wire, EDISON uses Power Line Carrier (PLC) communication to support the exchange of information. Although not part of IEC 61851, this technology is a valid candidate for future standardization.

An EDISON I/O board is installed at the EV and the EVSE and includes a PLC adapter.

To connect the EVSE with the aggregator, the IEC 61850 standard is used. IEC 61850 was originally aimed at substation automation, but has been expanded to cover the monitoring and control of distributed energy resources. The standard includes a reusable data model which can be used to monitor and control both the EVSE and the EV.

The EVSE implements an IEC 61850 compliant server which uses HTTP/HTTPS based RESTful web services, instead of the MMS protocol traditionally associated with IEC 61850. The REST interface is combined with the SIP application level protocol to better facilitate scalability. The use of Transport Level Security (TLS) is used to provide data confidentiality.

After the aggregator has extracted information through the REST interface of the IEC server it uses a software platform called the EVPP (EDISON Virtual Power Plant) that, through prediction and optimization, will compute a suitable charging strategy. The strategy is described using an IEC 61850-7-420 Energy and/or Ancillary Services Schedule (DSCH) which is sent to the EVSE IEC server. The schedule consists of a set of power set points and timestamps which will be followed by the EVSE during the charging of the EV.

An iPhone App and a webpage have been developed for the user to define charging requirements.

B. Vehicle to Grid technology, University of Delaware Vehicle-to-Grid (V2G) technology [11]-[12] is researched

and developed at the at the University of Delaware (UD). The

research focuses on the potential that V2G technology has for improving the utilization of the EVs as active resources in the grid and market. The research done by UD V2G spans a broad set of disciplines such as soft- and hardware development, grid impact and driving pattern analysis and aggregated fleet optimization. In addition to the technical aspects, UD V2G also covers policies, standards, legislation and user adoption. For testing and demonstrating, UD V2G uses a fleet of V2G enabled vehicles.

UD V2G is both a project and a research program and will continuously work within research, development and commercialization of V2G. Recently the group has stating replacing the term 'V2G' with 'Grid Integrated Vehicle' to emphasize the importance of grid integration.

AC Propulsion, a manufacturer of battery and propulsion systems, is an active partner in the project. The research is supported by the US Department Of Energy (DOE) as well as several American utility companies.

More information is available at www.udel.edu/V2G.

Fig. 2. The architecture of UD V2G

1) Market integration and EV utilization concept The UD V2G market integration concept has been tested by

participating in the regulation services market where the vehicle responds to regulation power requests sent from an Transmission System Operator (TSO). The Bi-directional charging utilization concept will allow the EVs to react to TSO requests for both up and down regulation and the EV User will be economically compensated for such services [13]. Regulating services has been implemented by UD V2G since it represents one of the most profitable markets to participate in. UD V2G publications have noted that their control mechanisms are also designed for other TSO markets such as spinning reserves, and for distribution system services such as peak load reduction, valley filling, reactive power, and transformer upgrade deferral, but UD V2G cars are not actually participating in these markets yet. 2) Architecture

The UD V2G architecture depicted in Fig. 2 can be classified as partly distributed since the EV implements an intelligent agent that will use a negotiation-like communication towards the aggregator. The EV will control the charging

4

process and be responsible for predicting and satisfying the energy requirements of the EV user.

Adding local intelligence and control in the EV can supply a better separation of concerns where a 3rd party, like the aggregator, would not have full control over charging and free access to utilization data. This secures the EV against external mismanagement (e.g. driver’s need for driving range has priority) and simplifies the optimization in the aggregator.

The purpose of the EVSE in the UD V2G architecture, aside from facilitating the power supply and internet connection, is to supply information on possible grid-related charging constraints. The user can through a web interface formulate her or his driving requirements. The collected trip information is then feed to the vehicle. 3) Technologies and standards

In the UD V2G project setup, an electric vehicle contains a Vehicle Smart Link (VSL) implemented on a automotive-grade Linux computer. The VSL will communicate with the Vehicle Management System (VMS) and Battery Managements System (BMS) of the EV to get battery information and to control charging. The Society of Automotive Engineers (SAE) J1772 standard is used for the equipment connecting EV and EVSE. J1772 defines the electrical and physical characteristics of conductive charging for EVs including requirements to inlets and connectors. Either in-band communication over the pilot line, or Power Line Carrier (PLC) communication is used for sending data via the charging cord.

Knowledge of grid and EVSE constraints will be captured by an XML file residing in the EVSE and will be sent to the VSL upon plug-in. This and direct wired communications ensure that the vehicle knows its electric power system node location, and the constraints, billing, and allowed services at that point. The EVSE is a repository of information (in the XML file) rather than a computing agent with active control over V2G.

The VSL will communicate the battery status and trip predictions to the aggregator to signal the capacity of each EV. The aggregator uses the UD V2G 'Coalition server' software to calculate the capacity of the EV (including grid constraints, battery state of charge, and scheduled or anticipated driver needs) and dispatch the TSO regulation requests accordingly. The result is a stream of 'power requests' from the aggregator to the vehicle that states the number of watts with which the EV should charge or discharge.

Both the VSL and the aggregator use software agents based on the Java Agent Development Framework (JADE). The agents communicate via the Agent Communication Language (ACL) as defined by Foundation for Intelligent Physical Agents (FIPA). The concepts of coalition formation and multiagent systems are applicable to the project in the sense that multiple entities (EVs) will group and cooperate to achieve some common goal, in this case revenue through TSO services. The UD V2G architecture, however, differs from the

above concepts in that a coordinating entity, namely the aggregator, will be present.

C. E-Mobility Berlin pilot project The German E-Mobility Berlin [14] project was initiated by

Daimler AG (Mercedes-Benz) and the utility RWE. Among the participants are also battery, EVSE and other automobile OEMs. The project introduces a fleet of 100 EVs supplied by Daimler and 500 EVSE’s which are delivered and powered by RWE in the streets of Berlin for a large field test. The project is aimed at developing and testing standardized solutions for electric vehicles.

Daimler is heavily involved in EV standardization in Europe and supports the vision of roaming. The project was launched in 2008 and will continually expand the field tests with new vehicles and technologies. More information is available at www.rwemobility.com.

Fig. 3. The architecture of e-mobility Berlin

1) Market integration and EV utilization concept The aggregator in the e-mobility project is initially seen as

the utility company e.g. RWE, who could sell energy to EV users and reward them for flexibility. By letting the user specify an 'end of charge' time the utilization concept of smart charging is supported. E-mobility does not directly address bi-directional charging and the use of the EV for ancillary services. The protocols and use cases of the project, however, is designed to be open for additional unspecified utilization concepts. 2) Architecture

The e-mobility project puts a lot of emphasis on the EV-EVSE interaction in its architecture. As illustrated in Fig. 3 the EV acts as a client towards a server implemented at the EVSE. The tariffs and charging options of the utility company will be represented by the EVSE which will serve as a proxy. Despite the presence of an aggregating entity, the setup is only partly centralized since the EV will implement a lot of decision logic by knowing the needs and requirements of the end-user and use them in a negotiation-like communication with the EVSE. 3) Technologies and standards

Each car is equipped with a Smart Charge communication unit which can communicate with the EVSE by using the e-mobility Smart Charge Protocol (SCP) over PLC. The SCP

5

defines a series of application level messages which are sent back and forth in the following sequence. After a plug-in has been detected, 'identification' messages will be used to configure the connection session and to establish identification, billing and contract details (for roaming). The EV will then request a list of EVSE provided services in a 'service discovery' message. Services include the charging and payment options available at the specific EVSE. The EV will then send its energy demand and intended charging behaviour in a 'power discovery' message. The EVSE will compare the charging behaviour with knowledge on local grid and equipment capabilities and send back price listings. When charging and billing has been settled, a series of messages initiates the power connection and monitors the charging process. SPC messages are encoded according to the Smart Message Language (SML) which is a mark-up language similar to XML that has been used for smart meter communication. Transport Layer Security (TLS) is used to supply data confidentiality through encryption. The DoIP protocol is used for EV diagnostics.

The e-mobility project has contributed significantly to the standardization of the IEC IEC 62196-2 Type 2 compatible Mennekes plug. The EVSE equipment supports conductive charging in accordance to IEC 61851.

III. PROJECT COMPARISONS AND RECOMMENDATIONS The investigation of the three EV integration projects has

revealed a series of differences and similarities which can be used to identify the areas that should be addressed by standardization. Fig. 4 compares the three projects based on the parameters described in the previous sections.

Fig. 4. Project comparison overview

A. Comparison - Market integration and EV utilization concept

Markets integration concepts for EVs will most likely vary based on local market conditions and are, for the centralized architectures, tightly linked to the business cases chosen by the aggregator. Smart charging, as covered by EDISON and e-mobility, can be seen as a first approach to market integration by allowing the EV to respond to variable energy prices on the energy market. The V2G equipment tested by UD V2G can be seen as the second step where the EV increases its potential as

a resource to the grid and more market integration concepts becomes possible. New integration concepts might occur over the coming years as markets will adjust to better handle distributed energy resources as part of the smart grid concept. By letting standards support as many utilization concepts as possible, it is ensured that a broad set of integration concepts will be possible in both present and future markets.

B. Comparison - Architecture The differences in the architectures of the three projects are

an indication of a fundamental question in EV integration: Which entity is responsible for influencing or controlling the EVs utilization behaviour? Is it the aggregator, the EVSE or the EV itself? Both e-mobility and UD V2G uses an EV-centric approach where an embedded computer in the vehicle will take certain decisions. Both the centralized and distributed architecture comes with a set of advantages. Theoretically, however, whether the controlling logic is placed at the aggregator or in the individual EV should not matter as long as optimization objectives are the same and EV User SLA's are strictly followed. One solution could be to have the infrastructure support a set of standardized architectures so that, for instance, both 'self managing' and 'non self managing' EVs can charge using the same equipment.

C. Comparison - Technologies and standards As can be seen on figure 4, EDISON is the only project to

use a standardised communication approach between EV and aggregator, namely IEC 61850. There are advantages to using an approach as well-described and widely applied as IEC 61850 and initial tests by EDISON indicates that the standard either supports, or can be extended to support, the utilization concepts described in this paper. A possible drawback of IEC 61850 is that it offers a structural decomposition of equipment as it is typically seen in SCADA applications. Technical de-composition means that the data model of IEC 61850 is designed to precisely represent physical components (inverters, chargers, relays etc.) and their operational status values. This can be argued to only make sense in a fully centralized architecture where direct and detailed access to equipment is appropriate. New standards will appear that ultimately might replace IEC 61850. Both e-mobility and EDISON have partners involved in the standardization around 'ISO/IEC 15118 – Vehicle to grid communication interface' which could have a major influence on future EV communication. On the SAE side J2836, J2847 and J2931 are addressing similar issues. Standards that describe the equipment used for conductive charging are used in all three projects. The IEC 61851 and IEC 62196-2 standards used in the European projects are similar to SAE standards used in the American UD V2G project. Adoption of these standards bodes well for roaming support. Especially Daimler in the e-mobility project is deeply involved in both IEC, ISO and SAE standardization work.

6

These efforts could help bridge the gap between European and American EV standards. The projects all use the internet and the associated IP stack in the communication solution as well as PLC on the physical level. The UD V2G group claims that wired communication is required to precisely identify the EVs location on the grid. Wireless communication, however, might become standard in EVs as new features and services become popular and it is claimed that 4G will have a major role in smart grids in general. If the localisation issues mentioned by UD V2G can be overcome, Wireless technologies should be considered a potential candidate for future EV communication standardization. While the projects are aligned on the lower parts of the OSI stack, they are very divided on the application level protocols. The solutions span from the structural decomposition used by IEC 61850 to the use of agent software. Application level protocols for EV communication must be aligned, or at least compatible, to support roaming above the equipment level.

All projects use the transport level security (TLS) cryptographic protocol as part of the communication stack. This is a well tested protocol used, for instance, in internet banking. This protocol alone, however, does not address all cyber security issues that the ICT architecture will face. It is recommended to align EV communication security with smart grid related security standardization work of IEC TC57 WG 15 (e.g. IEC 62351) and NIST (e.g. 'Smart Grid Cyber Security guidelines')[15].

IV. CONCLUSION The following are the key findings and recommendations by

this paper based on the project comparison:

• Market integration - Present and future market integration concepts will improve the utilization of EVs as active resources in the power market and system. Market integration concepts could take many forms and should ultimately be left to the businesses driving the aggregators. Equipment and communication standards should support as many utilization concepts as possible, so that the EV may be used in as broad a range of market integration concepts.

• Architecture - The placement of control logic and responsibilities between EV, EVSE and aggregator, differs between the integration projects investigated. One key question is how much intelligence will implemented at the EV. If a common architecture is not found, then standardization efforts must respect and manage the resulting diversity.

• Communication standards - A new class of standards covering EV communication, such as ISO/IEC 15118, are emerging. These standards should ideally support a broad set of utilization concepts. If, however, vehicles and EVSEs will be built to support very different architectures, one common solution is less feasible.

• Equipment standards - The Projects in question aligns nicely with the European and American standardization work - This trend brings great promise for roaming on the equipment level.

• Communication protocols - While the use of the internet protocol stack on top of PLC is used in all projects, application layer protocols differs significantly. These protocols should be aligned, or at least be compatible. Wireless communication for EVs is not used in any of the projects but could be considered a valid candidate for future work.

• Security - Transport level security, as used by all three projects, does not address all security concerns, but is a well-tested internet protocol suitable for handling confidentiality. Standardization aimed at smart grid security in general is also applicable for EV integration. Existing standards and guidelines by IEC and NIST should be considered.

If the above recommendations are considered, it will help promote a standardized platform for EV interoperability and grid integration. Such a platform can help realise two different potentials. The first is roaming that can help better servicing of driving requirements of EV owners and improve the utilization of infrastructure. Another potential is that most home charging equipment and EVs, across a series of different OEMs, will support smart grid integration and smart charging. Realising both of the above potentials will substantially aid and promote the cause of electric transportation.

V. REFERENCES [1] O. Sundström and C. Binding, “Optimization methods to plan the

charging of electric vehicle fleets,” in Proc. 2010 ACEEE International Conference on Control, Communication and Power Engineering, pp. 323-328.

[2] K. Clement, E. Haesen, and J. Driesen, "Coordinated charging of multiple plug-in hybrid electric vehicles in residential distribution grids," in Proc. Power Systems Conference and Exposition 2009, pp. 1–7.

[3] 29.F. Tuffner and M. Kintner-Meyer, “Using electric vehicles to meet balancing requirements associated with wind power,” Pacific Northwest National Laboratory (PNNL), Richland, WA (US), Tech. Rep., 2011.

[4] J. M. Jørgensen, S.H. Sørensen, K. Behnke, P. B. Eriksen, "EcoGrid EU- A prototype for European Smart Grids," Energinet.dk, 2011, Available: http://www.eu-ecogrid.net/index.php/documents-and-downloads

[5] S. You, C. Træhold, and B. Poulsen, "A market-based virtual power plant," in Proc. 2009 IEEE International Conference on Clean Electrical Power (ICCEP).

[6] N. Hatziargyriou et al. "Mobile Energy Resources in Grids of Electricity: the EU MERGE Project" in Proc. 2010 2nd European Conference on Smart Grids and E-Mobility, pp. 14-21.

[7] Green eMotion, Project web page: www.greenemotion-project.eu [8] The EV Project, Project web page: www.theevproject.com/ [9] C. Binding, D. Gantenbein, B. Jansen, O. Sundstroem, P. B. Andersen,

F. Marra, B. Poulsen, C. Traeholt "Electric Vehicle Fleet Integration in the Danish EDISON Project – A Virtual Power Plant on the Island of Bornholm". in Proc. 2010 IEEE Power & Energy Society General Meeting. Also available as IBM RZ 3761.

[10] J. Østergaard, A. Foosnæs, Z. Xu, T. A. Mondorf, C. A.Andersen, S. Holthusen, T. Holm, M. F. Bendtsen, K. Behnke,

7

"Electric Vehicles in Power Systems with 50% Wind Power Penetration : The Danish Case and the EDISON programme" Presented at the European Conference Electricity & Mobility. Würzburg, Germany, 2009

[11] W. Kempton and J. Tomic,, "Vehicle-to-grid power implementation: From stabilizing the grid to supporting large-scale renewable energy," Journal of Power Sources, vol. 144, no. 1, pp. 280–294, June 2005.

[12] S. Kamboj, N. Pearre, W. Kempton, K. Decker, K. Trnka, C. Kern "Exploring the formation of Electric Vehicle Coalitions for Vehicle-To-Grid Power Regulation" in Proc. 2010 AAMAS workshop on Agent Technologies for Energy Systems.

[13] W. Kempton and J. Tomic, "Vehicle-to-grid power fundamentals: Calculating capacity and net revenue" Journal of Power Sources, vol. 144, no. 1, pp. 268–279, June 2005.

[14] E-Mobility Berlin project, RWE project description: https://www.rwe-mobility.com/web/cms/de/1158022/emobility/

[15] NIST Smart Grid Cyber security guidelines, http://csrc.nist.gov/publications/PubsNISTIRs.html#NIST-IR-7628

VI. BIOGRAPHIES Peter Bach Andersen has a M.Sc. in Informatics and is currently finalizing a PhD at the Department of Electrical Engineering within the Technical University of Denmark (DTU). His research focus is on smart grids in general but with a special emphasis on electric vehicle integration. Peter is an IEEE student member.

Rodrigo Garcia received the electrical engineering degree and the MSc.degree in Electric Power Systems in Mexico, in 2001 and 2003 respectively. He obtained his Ph.D. degree from the University of Glasgow, U.K., in 2007. He holds the position as Assistant Professor at the Department of Electrical Engineering within the Technical University of Denmark (DTU). Among his research interests are dynamics, stability and control of electric power systems; renewable energy integration; modelling

and simulation of FACTS and custom power controllers; Synchrophasor Measurements; and Wide Area Monitoring, Protection and Control. He is the Secretary/Treasurer for the IEEE-PES Denmark Chapter. He is an IEEE, IET and CIGRE member.

Willett Kempton is Associate Professor in the Collage of Marine Studies, and Senior Research Scientist at the Center for Energy and Environmental Policy, both at the University od Delaware. He is conducting research and developing projects on both offshore wind power and V2G.

C Smartly Charging the Electric Vehicle Fleet

C Smartly Charging the Electric Vehicle Fleet

This chapter is part of the book “Smart Grid Applications, Communications, and

Security”, Lars T. Berger, Krzysztof Iniewski, Wiley, ISBN: 978-1- 1180-0439-5, pages:

381-408, 2012.

115


D Implementation of an Electric Vehicle Test Bed Con-

trolled by a Virtual Power Plant for Contributing to

Regulating Power Reserves

This paper was published in the Proceedings of the IEEE Power and Energy Society

General Meeting, 2012, San Diego, CA, USA.

144

1

Abstract-- With the increased focus on Electric Vehicles (EV)

research and the potential benefits they bring for smart grid applications, there is a growing need for an evaluation platform connected to the electricity grid. This paper addresses the design of an EV test bed, which using real EV components and communication interfaces, is able to respond in real-time to smart grid control signals. The EV test bed is equipped with a Lithium-ion battery pack, a Battery Management System (BMS), a charger and a Vehicle-to-Grid (V2G) unit for feeding power back to the grid. The designed solution serves as a multifunctional grid-interactive EV, which a Virtual Power Plant (VPP) or a generic EV coordinator could use for testing different control strategies, such as EV contribution to regulating power reserves. The EV coordination is realized using the IEC 61850 modeling standard in the communication. Regulating power requests from the Danish TSO are used as a proof-of-concept, to demonstrate the EV test bed power response. Test results have proven the capability to respond to frequent power control requests and they reveal the potential EV ability for contributing to regulating power reserves.

Index Terms— Electric vehicles, Test Bed, Regulating power, Virtual Power Plant

I. INTRODUCTION HE Electric Vehicles and Plug-in Hybrid EV (PHEV) are expected to play an important role in the future power

system. Within smart grids research, electrical transportation has a complementary role in the overall system management of energy and power [1]; moreover the European target on reducing CO2 emissions and increasing penetration of renewable energy are among the major drivers for the research [2].

The energy storage capability is the key factor for smart grid applications of EV in power system. When parked and plugged into the grid, EV are expected to either charge intelligently, or discharge feeding power back to the grid [3]. In the latter case, EV would enter a mode known as Vehicle-to-Grid (V2G), permitting the provision of several grid

This work was in part funded through the EDISON project, a research

project of the public Danish transmission system operator Energinet.dk’s research program FORSKEL, (Project Number 081216).

F. Marra, D. Sacchetti, A. B. Pedersen, P. B. Andersen, C. Træholt, and E. Larsen are with the Centre for Electric Technology, Department of Electrical Engineering, Technical University of Denmark, 2800 Kgs. Lyngby, Denmark (e-mail: [email protected]).

services [4]. In general, if the individual EV can be intelligently managed, a large number of such vehicles can become an asset in the future power system. The charging process could be controlled by modulating the charging power, as well as the discharging process, by enabling the V2G mode when there is a need from the grid [5]. Many projects are addressing the aforementioned EV operation, as coordinated charging using different simulation tools. In [6], the authors analyze, through dynamic simulations, the potential daily profits for EV users, with the provision of regulating power. In [7], the authors studied the benefits offered by EV for facilitating the integration of large scale wind power in Denmark; EV fleets are modeled in a simulation platform as storage units when charging or as small generators during V2G operation. Galus et al. in [8] presented a method for tracking secondary frequency control using groups of PHEV and a simulation platform to simulate an EV aggregator.

The participation of EV in regulating power schemes is possible using an aggregation entity for EV coordination. This is done in the Danish EDISON project [9]-[10], where the contributors proposed a centralized coordination solution for an efficient integration of EV in the power system. The aggregation technology is based on the Virtual Power Plant (VPP) concept [11], where the Edison VPP is the EV coordinator.

Evaluating the contribution of an EV for regulating power reserves in a VPP framework, where a huge amount of communication and hardware interfaces are involved, gives raise to the need of new grid-interactive evaluation platform.

This paper describes the implementation phases of an EV test bed, working under the coordination of a VPP and contributing in regulating power reserves, as secondary frequency control [12]. A real regulating power request from the Danish Transmission System Operator (TSO) is processed by the VPP and sent in form of a charging/discharging power schedule to the EV test bed.

Test results performed using the EV test bed show that an EV is in fact capable of real-time communication with a VPP and can quickly react to contribute to grid power reserves.

II. ELECTRIC VEHICLES FOR SMART GRIDS

The interaction between EV and the electric power system is only possible if the vehicles can connect to the electrical

Implementation of an Electric Vehicle Test Bed Controlled by a Virtual Power Plant for

Contributing to Regulating Power Reserves F. Marra, Student Member, IEEE, D. Sacchetti, A. B. Pedersen, P. B. Andersen, C. Træholt,

and E. Larsen

T

2

grid for charging. An effective interaction between EV and grid requires the

combinations of different factors such as:

- Grid interactive vehicle architectures; - Controllable charging/discharging operation;

A. Electric vehicle architectures Among different types of hybrid electric vehicles and pure

electric vehicles, a general distinction is based on their ability to plug-in. In this work, non plug-in hybrid vehicles will be disregarded, as an interconnection with the grid is not possible.

This section lists the system architectures capable of charging using the grid [13]. There are mainly two classes of plug-in EV: plug-in hybrid EV (PHEV) and battery-powered EV. An overview on the different architectures is depicted in Fig. 1.

Fig. 1. Plug-in Electric Vehicles system architectures. (a) Series-parallel hybrid EV. (b) Series hybrid EV. (c) Parallel hybrid EV. (d) Battery-powered EV

In the PHEV class, three variants have been developed so far:

- Series-parallel hybrid - Series hybrid - Parallel hybrid

The main difference among the topologies is the drive system used and the interconnection of its components, before the power is transferred to the wheels.

In the series-parallel hybrid vehicle, Fig. 1 (a), the system is designed to operate both in a series or parallel configuration. The reconfigurable system is made possible by the use of a planetary gear, which is the mechanical coupling (MC) for the three machines. In the series hybrid vehicle, Fig. 1 (b), the electric traction system and Internal Combustion Engine (ICE) system operate in a series connection. In sequence, the ICE is coupled with a generator (Gen.) which generates the electric power for recharging the battery, the battery then supplies an electric motor driver to transfer power to wheels. In the parallel hybrid vehicle, Fig. 1 (c), the ICE and electric motor (EM) operate in parallel mode, where the ICE supports the electric traction at certain points of the driving pattern, e.g. when higher power is needed to the wheels.

In the battery-powered EV class, Fig. 1 (d), the drive system is realized using only an electric motor and a motor driver. Therefore the only energy source is the battery pack.

In this work a battery-powered EV is the architecture chosen for the EV test bed implementation.

B. Controllable charging/discharging operation All plug-in EV are able to absorb power from the grid while

charging their battery packs. The controlled charging or discharging (V2G) process can be achieved using different infrastructure concepts, such as home charging or public charging stations [14]. According to the IEC 61851 standard [15], the most common power rates for domestic and public charging are depicted in Table I.

TABLE I CHARGING POWER RATES

AC current AC voltage Grid connection Power 10 A 230 V single phase 2.3 kW 16 A 230 V single phase 3.7 kW 32 A 230 V single phase 7.4 kW 16 A 400 V three-phase 11 kW 32 A 400 V three-phase 22 kW

All power rates, regardless of charging or discharging, are characterized by an AC current, usually 16A or 32A, and based on the grid connection type, single-phase or three-phase.

In this work a charging/discharging power rate of ± 2.3 kW is used for the experimental validation.

C. Planned EV test bed operation with Virtual Power Plant The EV system architecture is planned to respond to

different control signals from a centralized EV coordinator. The control signals for the vehicles can be generated by a

VPP and based on different variables such as the power system frequency, the market spot price and others.

In this paper, the centralized control concept for EV fleet management as described by Binding et al. in [10] is used as a study framework. The EV test bed operation is planned within the VPP framework depicted in Fig. 2. Different interfaces have been defined to establish communication between the Edison VPP and other entities in the architecture. While a

3

generic VPP can aggregate and control various distributed energy resources (DERs), e.g. combined heat and power units (CHPs), PV plants, wind turbines, medium/large consumers, other power units and smart houses, in this paper, only EV are considered.

An interface with the TSO is defined to receive the activation commands for accepted regulating power reserves contracts.

An interface with the Distribution System Operator (DSO) is also defined to collect the grid status for the location of every connected EV. Grid constraints are considered at this interface, to ensure that the charging/discharging operation complies with power quality issues. In addition, the metering information for accounting is also collected via the DSO interface.

Fig. 2. EV test bed operation in a Virtual Power Plant framework

The transaction interface with a billing provider is used to

perform billing to the resources providing regulating services. In this paper, the EV-VPP interface is implemented to

establish communication between the VPP and the EV test bed. The control requests for the EV test bed are generated from the VPP, based on the grid needs of regulating power reserves of the TSO.

III. TEST BED IMPLEMENTATION Designing an EV test bed for testing the potential EV

operation with a VPP was performed in two phases:

• Planning the EV test bed architecture • Dimensioning the EV components

A. Planning the EV test bed architecture With reference to the EV architectures described in Section

II, a battery-powered EV architecture was the choice for the EV test bed implementation. The main reasons for choosing this architecture is that with a pure battery EV, zero emissions can be achieved during driving [16], while grid interaction is more meaningful, due to a larger storage capacity.

In a battery-EV the following components can be considered:

- a battery pack - a battery charger - a BMS - a three-phase motor driver - an electric motor

For the scope of the study, the three-phase motor driver and the electric motor are not needed, therefore these two components were not considered in the development.

Since with the architecture of Fig. 1 (d), the V2G operation is not possible, this was enhanced by adding a V2G unit which could operate in a complementary way to the charger. The implemented EV test bed architecture is shown Fig. 3.

Fig. 3. EV test bed architecture

The battery pack is interfaced to a battery management system (BMS), which monitors its status.

The charger is designed as an AC/DC converter, directly connected to the main grid, by means of a three phase cable connection; while the V2G unit is made of a single phase DC/AC inverter.

It is worth noticing that, since the aim of the EV test bed is to emulate a real EV connected to the grid, all EV components were dimensioned according to realistic EV energy and power levels.

B. EV test bed components The design of the battery pack took into account the

following requirements:

• Common designs of battery-EV [17][18] • V2G operation requirement

The choice of a battery technology to use for the test bed was based on the analysis of current market trends for EV. Some consulting companies, e.g. Frost & Sullivan [19], foresee more than 70% of EV in 2015 to be powered by lithium-ion (Li-ion) batteries. Compared to other battery technologies, Li-ion batteries offer a greater energy-to-weight ratio, greater power levels and low self-discharge when not in use [16]. For the reasons mentioned, a Li-ion battery was the choice for the EV test bed.

The electrical features of the battery pack were chosen considering common designs of EV battery packs. Generally

ba40pabaVpno

is

cowa4 aroint

baopopVPtem oufor

Thful

is

attery-EV have00V and a batteack was designattery cells, whpack, of 363 V. Eominal capacity

Based on the valid for calcu

E

where N is the

The requiremonverter with inas used. The rakW, which leound 16 A. Tternal transformA battery m

attery pack. Theperation of thperation. It estiPP and monmperature. The charger w

utput voltage rmula shown b

dcV

he current Idc oll load, which lThe implemendepicted in Fig

e battery pack ery capacity oned integratinhich leads to Each cell has ny Cn, of 40 Ah.nominal param

ulating the nom

n pack nE V C= ⋅

e number of cel

ment of V2G nput DC voltagated output powads to a maxi

The V2G invemer, which serv

management sye main functionhe battery paimates the SOCnitors the b

was designed Vdc was dim

below, accordin

1.25 packV= ⋅

on the battery leads a charginnted test bed wg. 4.

Fig. 4. E

voltages in thf at least 10-1g 110 Li-ion a total nomin

nominal voltage

meters, the follminal battery en

n n nN V C= ⋅ ⋅

lls.

operation, age in the rangewer of the V2Gimum generateerter is also eves as galvanic

ystem (BMS) n of the BMS i

ack during chC information battery voltag

as a single-phmensioned usinng to [20]:

453 V=

side is dimensng power of abwith integrated

V test bed

he range of 35 kWh. A battseries connec

nal pack voltae Vn, of 3.3 V a

owing expressnergy:

n (1)

a DC/AC powe of 250 – 500

G inverter is abed AC currentequipped with c isolation. is linked to is to ensure a sharging or Vwhich is used

ge, current a

hase AC/DC. Tng the empiri

(2)

sioned of 10 Aout 4.5 kW.

d EV compone

00-tery cted age, and

sion

wer 0 V bout t of

an

the safe

V2G d by and

The ical

A at

ents

IAs pr

operate the direpurposeadapt thcontrol

F

To finterfacecommun

The 61850 sattempt system to HTTP

The Val. [23] schedulepower re

Thouare simpassociatapplicab[24].

IV. COMMUNI

reviously mentas part of a ce

ect control of e of future resehe software syschemes.

Char

Inverter

PowerMeasuremControl

Fig. 5. EV test bed

facilitate cente was impnication and cocommunicatiostandard [22].to promote thecommunicatioP/REST. This wVPP used for and has been

es, in responeserves specifi

ugh used for anply based on pted time stampble to any type

ICATION AND Ctioned, the EVentralized aggr

a VPP, as dearch, it will bystem in order

V

IECS

IEC

B

B

rger

relay

ment

d communication

tralized coordlemented in ontrol architecton is based on. As an acadee use of existin

on, the IEC 61work was presthis paper wasused for gene

nse to the reqied by the TSO

n EV test bed inpositive/negativp. For this ree of Distributed

CONTROL WITH

V test bed was regation frame

described in [2be possible, in r to e.g. test d

VPP

C 61850 Server

C Module

BMS

Battery

relay

and control archite

dination, the accordance

ture depicted inn the well-estaemic exercise ng web standar850 standard w

sented in details designed by erating and senquirements for

O. n this case, theve power requason, they ared Energy Reso

4

VPP designed to

ework, under 21]. For the any case, to

decentralized

ecture

VPP-to-EV with the

n Fig. 5. ablished IEC

and in the rds in power was mapped s in [23]. Pedersen et

nding power r regulating

se schedules uests with an e potentially ource (DER)

5

A. IEC 61850 Server and Module The server, which was developed in compliance with the

IEC 61850 standard, is designed based on a modular plug-in architecture, in order to facilitate an easier adaptation and installation of new devices of virtually any type [23].

A device specific plug-in, or IEC Module, was implemented, in order to enable direct control of the test bed from the VPP, as well as to facilitate the collection of battery status information along with any other measurements.

Because the charger and inverter are two separate pieces of hardware connected to the same battery, an algorithm was written into the plug-in module to guarantee the mutually-exclusive operation of the connecting relays. This prevents the simultaneous operation of both devices.

Another communication link is established between software plug-in and the BMS to extract the state-of-charge (SOC) information from the battery.

B. Charging As indicated by Fig. 6, there is no direct control link

between the IEC Module and the charger; this is because the charger has not communication interface for remote control. For the purpose of this paper, the charger is fixed to a fixed power rate and is coupled or decoupled from the battery by means of a DC relay, which is controlled via RS232, as illustrated in Fig..

IEC Module

Relay

RS232

Charger

Grid

Fig. 6. Charging communication in detail

As previously mentioned, the charger and inverter are both connected to the battery pack using two mutually exclusive relays. In order to ensure that the devices have exclusive “access” to the battery, a simple timing scheme was used in the IEC Module. As depicted in Fig., a time gap was added between two switching events to ensure a safe transition.

Fig. 7. Relay timing scheme

C. V2G The coupling of the V2G inverter is achieved by means of an

identical DC relay as for the charger. Under V2G operation, the generated power level is controllable and this is managed through an attached communication hub. The same hub implements an HTTP/JSON web interface. A more detailed illustration of the V2G architecture used for the EV test bed is depicted in Fig. 8.

Fig. 8. V2G communication in detail

D. Battery Status Information The real-time status of the battery is monitored by the BMS.

The BMS information is acquired by the IEC Module using RS485 based serial communication link. By means of this link, all battery data can be extracted as a set of values and made available to the VPP via IEC 61850 for detailed monitoring. The set of values includes:

• Battery pack voltage • Current • State of charge • Temperature in different areas of the battery pack • Remaining energy and single cell voltages

Fig. 9. BMS communication in detail

The BMS comes equipped with an RS232 port for remote monitoring of the battery pack, as well as controlling various limits/alarms. Connected to the BMS are a series of sensors, which connection are depicted as a series of smaller wires in Fig..

6

Fig. 10. Test results of EV test bed contribution in regulating power service – Secondary control. (a) Target regulating power of the Danish TSO and VPP

generated schedule for the EV test bed and the other EV. (b) Measured power response and battery SOC of EV test bed.

V. EXPERIMENTAL RESULTS The EV test bed participation in regulating power reserves

was tested considering the regulating power required for secondary control, by the Danish TSO, on the 1st of January 2009 [25]. The regulating power profile sent from the TSO to the VPP is given in 5-minutes average MW values as shown in Fig. 10(a). The target regulating power is derived by the sum of all up and down regulation requests sent out by the TSO to an array of providers in the same 5-minutes. The target anyway does not reflect the exact need of the system but rather the value is used to drag the providers in the right direction.

Nevertheless, the target value is a very good approximation to the real-time need of regulation reserves. A new schedule, related to the TSO power target, was generated by the VPP every 5-minutes, and sent to the EV test bed. The regulation was tested in the time interval 16h30 to 19h30. The schedule is shaped as ±2.3 kW power requests with time stamps, indicating the activation/deactivation time of charging and V2G mode, Fig. 10(a). Since each EV has a very small capacity compared to the grid needs, it was assumed that the VPP meets the TSO target by aggregating a number of simulated EV. The EV system response, Fig. 10(b), is taken measuring the electric power flow at the point of

-2.5

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

-100.0

-80.0

-60.0

-40.0

-20.0

0.0

20.0

40.0

60.0

80.0

100.0

Sche

dule

(KW

)

Targ

et V

alue

(MW

)

Targetvalue (MW) Schedule (KW)Target value(MW) Schedule (kW)

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

-3-2.5

-2-1.5

-1-0.5

00.5

11.5

22.5

3

SOC

Mea

sure

d Po

wer

(kW

)

ActivePower SOCMeasured Power (MW) SOC

7

common coupling (PCC) of the EV test bed with the grid. The power was recorded with 1 minute sampling time. Test results are depicted in Fig. 10(b). It is possible to observe that the EV test bed is able to react in real-time to the power schedule sent by the VPP. The measured power profile in Fig. 10(b) validates the effectiveness of the EV architecture proposed in this work. Furthermore, the SOC profile shows the energy variation in the EV test bed battery, during the regulation service. The EV test bed started to contribute to regulating power service with an initial SOC of about 0.5 or 50%.

VI. CONCLUSIONS Testing the capabilities of EV for smart grid applications

requires the development of adequate evaluation platforms. In literature it was demonstrated that EV can potentially operate under a number of coordination schemes, including the participation in regulating power reserves. While this was extensively presented by simulation scenarios, in this paper, a real implementation of regulating power reserve performed by a full-scale EV test bed was presented. The test bed was designed to flexibly interact in real-time with an EV coordinator and the electricity grid, under different coordination concepts. To do so, real EV components and communication interfaces were used, that make possible an end-to-end interaction with a VPP. The implementation of an EV test bed from scratch enabled the management of the single components involved in the EV system: charging/discharging units and BMS. With the implemented communication and control architecture it was possible to establish a stable communication between the EV the test bed and the Virtual Power Plant. The potential offered by EV for regulating power was demonstrated testing the EV test bed hardware and software interfaces. An array of regulating power requests (load frequency control) within a 3-hours time interval, sent by the Danish TSO on the 1st of January 2009, was used as study case. The TSO target values were converted to an EV compatible schedule by the Edison VPP and sent down to the EV test bed among the other simulated EV. Test results revealed the potential capability of EV to respond in real-time to different charging/discharging requests based on different coordination plans. Further investigations will be performed for evaluating the reliability of the communication involved, when several fleets of EV are simultaneously coordinated.

ACKNOWLEDGMENTS The authors would like to thank Sammarco Elettronica,

Italy, for the technical support offered in the realization of the Electric Vehicle test bed.

REFERENCES [1] J. A. Pecas Lopes, F. Joel Soares, and P. M. Rocha Almeida,

“Integration of Electric Vehicles in the Electric Power System”, in Proc. IEEE, Vol. 99, 2011

[2] European Commission, “20 20 by 2020 Europe's climate change opportunity”, COM (2008) 30 final, Brussels, 2008.

[3] A. Brooks, “Integration of electric drive vehicles with the power grid – a new application for vehicle batteries”, in Battery Conference on Applications and Advances. The Seventeenth Annual, 2002.

[4] W. Kempton and J. Tomic, “Vehicle-to-grid power fundamentals: Calculating capacity and net revenue”, J. Power Sources, vol. 144, no. 1, pp. 268–279, Jun. 2005.

[5] S. Deilami, A.S. Masoum, P. S. Moses, M. A. S. Masoum, “Real-time coordination of Plug-in Electric Vehicle Charging in Smart Grids to Minimize Power Losses and Improve Voltage Profile”, in IEEE Transactions on Smart Grid, Vol. 2, no. 3, pp. 456-467, 2011.

[6] N. Rotering, M. Ilic, “Optimal Charge Control of Plug-in Hybrid Electric Vehicles in Deregulated Electricity Markets”, in IEEE Transactions on Power Systems, Vol. 26, no. 3, pp. 1021-1029, 2011.

[7] J. R. Pillai and B. Bak-Jensen, “Vehicle-to-grid systems for frequency regulation in an Islanded Danish distribution network”, in IEEE Proc. Vehicle Power and Propulsion Conference (VPPC), 2010.

[8] M. D. Galus, S. Koch, G. Andersson, “Provision of Load Frequency Control by PHEVs, Controllable Loads, and a Cogeneration Unit”, in IEEE Transactions on Industrial Electronics, Vol. 58, Nr. 10, 2011.

[9] B. Jansen, C. Binding, O. Sundstrom, D. Gantenbein, “Architecture and Communication of an Electric Vehicle Virtual Power Plant”, in IEEE Proc. First IEEE International Conference on Smart Grid Communications, 2010

[10] C. Binding, D. Gantenbein, B. Jansen, O. Sundstrom, P. B. Andersen, F. Marra, B. Poulsen and C. Træholt, “Electric Vehicle Fleet Integration in the Danish EDISON Project - A Virtual Power Plant on the Island of Bornholm”, IEEE Power & Energy Society General Meeting, 2009.

[11] N. Ruiz, I. Cobelo, J. Oyrzabal, "A Direct Load Control Model for Virtual Power Plant Management", in IEEE Transactions of Power Systems, Vol. 24, No. 2, pp. 959-966, 2009.

[12] J. L. Aguero, M. C. Berequi, F. Issouribehere, “Grid Frequency Control. Secondary frequency control tuning taking into account distributed primary frequency control”, in IEEE Proc. of Power and Energy Society General Meeting, 2010.

[13] K. Chen, A. Bouscayrol, A. Berthon, P. Delarue, D. Hissel, and R. Trigui, “Global modeling of different vehicles”, IEEE Vehicular Technology Magazine, 2009

[14] B. Lunz, T. Pollok, A. Schnettler, R. W. De Doncker, D. U. Sauer, “Evaluation of Battery Charging Concepts for Electric Vehicles and Plug-in Hybrid Electric Vehicles”, Advanced Automotive Battery Conference (AABC), 2009

[15] IEC 61851, Electric vehicle conductive charging system – Electric vehicles requirements for conductive connection of an a.c./d.c. power supply.

[16] P. Morrison, A. Binder, B. Funieru, C. Sabirin, “Drive train design for medium-sized zero emission electric vehicles”, 13th European Conference on Power Electronics and Applications, 2009.

[17] General Motors, “Chevrolet Volt – Quick reference sheet”, online [18] Nissan, “Overview of Nissan Leaf”, in News Releases,

www.nissan-global.com, 2010 [19] Frost & Sullivan, “M5B6-Global Electric Vehicles Lithium-ion Battery

Second Life and Recycling Market Analysis”, public report, 2010. [20] Thunder Sky, Li-ion battery user manual [21] EDISON project, “WP3 report D3.1”, www.edison-net.dk [22] IEC 61850 Part 7-420 [23] A. Bro Pedersen, E. Bragi Hauksson, P. Bach Andersen, B. Poulsen, C.

Træholt, and D. Gantenbein “Facilitating a generic communication interface to distributed energy resources”, in IEEE Proc. First IEEE International Conference on Smart Grid Communications, 2010

[24] A. Timbus, M. Larsson, and C. Yuen, “Active Management of Distributed Energy Resources Using Standardized Communications and Modern Information Technologies”, in IEEE Transactions on Industrial Electronics, Vol. 56, Nr. 10, pp. 4029-4037, 2009

[25] Energinet.dk, “Load Frequency Control - Secondary Reserve data”, January, 2009.


E Prediction and Optimization Methods for Electric Ve-

hicle Charging Schedules in the EDISON Project

This paper was published in the Proceedings of the IEEE PES Conference on Innovative

Smart Grid Technologies (ISGT), 2012, Washington, D.C, USA.

152

Prediction and optimization methods for electricvehicle charging schedules in the EDISON project

Andreas Aabrandt, Peter Bach Andersen, Anders Bro Pedersen, Shi You,Bjarne Poulsen, Niamh O’Connell and Jacob Østergaard

Dept. of Informatics and Mathematical Modelling & CET at Dept. of Electrical Engineering,Technical University of Denmark

Abstract—Smart charging, where the charging of an electricvehicle battery is delayed or advanced in time based on energycosts, grid capacity or renewable contents, has a great potentialfor increasing the value of the electric vehicle to the owner, thegrid and society as a whole.

The Danish EDISON project has been launched to investigatevarious areas relevant to electric vehicle integration. As part ofEDISON an electric vehicle aggregator has been developed todemonstrate smart charging of electric vehicles.

The emphasis of this paper is the mathematical methods onwhich the EDISON aggregator is based. This includes an analysisof the problem of EV driving prediction and charging optimiza-tion, a description of the mathematical models implemented andan evaluation of the accuracy of such models. Finally, additionaloptimization considerations as well as possible future extensionswill be explored.

This paper hopes to contribute to the field of EV integrationby coupling optimized EV charging coordination with the EVutilization predictions on which the former heavily relies.

I. INTRODUCTION

Besides offering a means of cleaned transportation, the elec-tric vehicle (EV) represents a high-powered, quick responding,controllable load that can adjust its charging behavior tothe day-ahead energy prices and even participate in ancillaryservice markets.

The Danish EDISON project [11],[12] has been launchedto explore the above potential by investigating topics relevantto EV integration and to demonstrate how the charging andpossible discharging of electric vehicles (EVs), if handledintelligently, can yield benefits to the EV owner, the grid andsociety

A wide range of different EV integration concepts arealready being investigated in numerous similar projects. In-tegration concepts typically differ in what needs the EV isused to fulfill in the power system and what mechanisms,designs and technologies are used to control or influence theEVs interaction with the grid.

The approach used in EDISON is having an Aggregatorentity control the interaction between an electric vehicle fleetand the electrical grid to facilitate ’smart charging’. TheEDISON aggregator is demonstrated both towards real EVsand a simulated fleet.

The aggregator software implemented in EDISON relies ona set of algorithms in order to predict the charging availabilityof the fleet and minimize the charging cost based on day-ahead energy markets. The main focus of this paper is on

the description and evaluation of a prediction method usingexponential smoothing. Besides minimizing the charging cost,a set of additional considerations imposed by the owner of thevehicle, the distribution grid and the EV battery should alsobe added to the optimization.

The remainder of the paper is organized as follows. SectionII will describe work by IBM and the University of Delawarerelevant to the EDISON project and this paper. In section IIIwill describe the mathematical problems that will be addressedby the models described in sections IV and V. Sections VIIand VIII will present results based on the empirical data used.Section IX will introduce regulation services on top of smartcharging. Finally, section X will summarize the findings of thepaper.

II. RELATED WORK

Sundstrom and Binding have, in a set of papers, focusedon coordinated EV fleet charging. In [13] the authors uselinear and quadric programming to minimize charging costs.In [8] the same authors propose an optimal charging methodwhere the aggregator is aware of the grid constraints. In athird paper [14] Sundstrom and Binding investigate how thecharging planning of an aggregator can be made to meet userrequirements.

This paper adds to these papers by focusing on the avail-ability prediction and by suggesting additional considerationsand constraints.

According to research by Kempton et al. from the UniversityOf Delaware (UD), an EV would be parked 95% of the timeand would thus represent a vastly underutilized resource if itspotential for servicing the grid is not used [15],[16]. From thesame research group, Kamboj has developed algorithms thatcalculate an electric vehicles ability to participate in frequencyregulating services while considering the energy requirementsand time constraints of the EV owner [17]. The EDISONAggregator software could borrow from the experiences of UDand there are plans to add regulating services to the objectivesof the EDISON aggregator software.

III. ANALYSIS OF THE PROBLEM

For an EV to adjust its charging behaviour according today-ahead energy prices and still meet the energy demandof the individual EV owner, mathematical models need to bedeveloped. A major concern is that reliable data of EV driving

Fig. 1. Meta model

behaviour is hard to come by. Some studies have been con-ducted but most of them are geographic and demographicallybiased.

The AKTA data [19] was originally collected in a collab-oration between the city of Copenhagen and the TechnicalUniversity of Denmark in order to investigate the drivinghabits of people living in the capital. The AKTA data, whichis used in this study, are collected by means of GPS wherea total of 360 vehicles were followed from 14 to 100 daysin 2001 - 2003. The AKTA data is biased towards drivingpatterns similar to everyday driving in the greater Copenhagenarea. The data is incomplete in that some observations aremissing. The data have been filtered to include only theobservations which have been plugged in during night timefor more than ten days during the period of the trial. Thereare 205 observations left in the sample after this filtration.

The present study is aiming at developing a robust frame-work not too sensitive to the errors of prediction. It isassumed that the day-ahead energy prices are known. It willbe investigated how to model the day-ahead demand of an EVowner and using the just stated estimation along with the day-ahead energy prices to calculate the optimal charging schedulefor a single EV.

The overall framework used in the present proceeding isdepicted in figure 1. At the individual level every EV will havea static energy demand until enough data has been collected.Once enough historic information is available a predictionmodel will be employed. Since mathematical models in gen-eral are sensitive to the topology of the data a simple predictionmodel will be developed. The prediction model and the day-ahead energy prices are fed to the optimization model at thelast expected plugin of the day which will produce a chargingschedule 24 hours ahead.

It is an advantage to let the prediction model and the opti-mization model have a low coupling. This makes it possible toplugin a new prediction model if a more appropriate model islater developed. This also allows for a static prediction modeluntil enough data has been collected.

IV. MATHEMATICAL MODEL FOR PREDICTION

For relatively small and biased data sets, simpler modelsare more suitable. If the model is very complex then it shouldtake into account the cyclic variations of the data, e.g. theremay be daily and weekly variations in driving patterns. Thesevariations may be different between geographical regions andculture. An exponential smoothing model is proposed whereno such variations are taken into account. This will be donesuch that the model can be homotoped between upper andlower bounds of the data. The ideal placement will be differentfor each EV.

Consider figure 2. The interval available for charging to bepassed to the optimization model is the period between themaximum of the plugin prediction interval and the minimumof the plugout prediction interval.

Fig. 2. Stopover duration.

An exponential smoothing model for estimating pluginintervals will be used. The model is given by the recursiveequation

Xn+1 = (1− δ)Xn + δSn, (1)

where δ ∈ [0; 1] is a fixed parameter, X1 = S1 and S ∈ Rn ishistoric data for a given vehicle.

Preprocessing the data to remove outliers will be necessaryfor sparse data. Let (Rn, ‖ · ‖) denote the usual n-dimensionalEuclidean space equipped with the usual norm, i.e

‖x‖2 =n∑

i=1

x2i , x ∈ Rn.

The variance of a sample x ∈ Rn is thus given by

σ2 =1

n− 1‖x− µ1‖2,

where µ is the mean of the sample and 1 is the identityelement in Rn. From now on the identity element will not bewritten when vector-scalar addition occur. Here x is interpretedas the time series of historic data for a given EV. It isassumed that behaviour which deviates sufficiently from themean is unpredictable and this leads to the following lemma,which states a criterion for an element in a sample to be”predictable”. This lemma will be used for variance reduction.

Lemma IV.1. Denote by µ and σ2 the mean and variance,respectively, of a sequence xn of positive real numbers. Letxnk be a subsequence of xn and let p be a fixed positiveinteger such that µ ≥ pσ. Then

{|xnk − µ| ≤ p σ, ∀k}m{xnk ∈ [µ− pσ; µ+ pσ] , ∀k}.

Proof: Since xnk , σ ≥ 0 for all k and x ∈ Rn it sufficesto consider

|xnk − µ|2 ≤ p2σ2 ⇔ x2nk + µ2 − 2µxnk − p2σ2 ≤ 0.

This is a quadratic inequality and solving it finishes the proofof the lemma.

Consider a sequence (xn) ∈ Rn with variance σ2 and meanµ. Let p ∈ N such that µ ≥ pσ. Then, for a subsequence(xnk) ∈ Rk, with variance σ2 and mean µ, satisfying

|xnk − µ| ≤ p σ, ∀k

then it is assumed that σ2 ≤ σ2.The above states that if a subsequence xnk is chosen such

that it is an integral number, p, of standard deviations σ awayfrom the mean µ, then the variance of this subsequence isbounded by the variance of the original sequence.

In [14], Sundstrom and Binding presents a linear pro-gramming model which uses time and energy margins tocompensate for prediction errors. The time and energy marginsare implemented by modifying the optimization constraints.

The method used here compensates for prediction errorsprior to optimization by considering the variance of thepredictions. This is done for time and could also be used forprediction errors of energy targets if energy was part of theprediction model.

After determining the stopover duration, the time periodwill be cropped in both ends depending on the standarddeviation of the prediction i.e. the more unpredictable thedriving behaviour, the more conservative the duration estimate.

V. MATHEMATICAL MODEL FOR OPTIMIZATION

A charging schedule defines the periods in time and charg-ing powers at which an EV should charge its battery. Theoptimization algorithm is the generator of the charging sched-ules.

Let α ∈ Rn+ denote the cost vector and write αi for the cost

at time interval i = 1, ..., n, where n is the number of timeintervals, and let the length of the time interval be βi, whereβ ∈ Rn

+. Let p ∈ Rm+ , such that pj denotes the power at which

the battery is charging, j = 1, ...,m, where m is the number ofpossible power configurations. Normally one could use the pvector as the decision variable, if {0} is included in the set ofpower configurations, but here it will be used that there existsa vector e such that pj =

∑pjej where ej ∈ Z2 = Z/2Z for

every j = 1, ...,m.The prediction model will be used to produce an estimate

of the lower and upper bounds for the number n.Let X ∈ Zn×m

2 , i.e. for every i it holds that Xi ∈ Zm2 .

Choose X as the decision variable and let f : Zn×m2 → R

be the cost function defined by multiplying every definedvariable in the obvious linear fashion. While the objective isto minimize the cost, the initial model to be used is then given

Fig. 3. Charging Schedule

by

min : fm,n =n∑

i=1

m∑

j=1

αiβipjXij

subject to :m∑

j=1

X1j ≤ 1

...m∑

j=1

Xnj ≤ 1

m∑

j=1

n∑

i=1

βipjXij ≥ E,

(2)

where E is the energy required by the EV. The product of theoptimization model is a charging schedule like the one shownin figure 3.

VI. VALIDATION OF RESULTS

The following example demonstrates the level of robustnessof the optimization model. This will also demonstrate that theprediction model need not be very precise, which is goodbecause it is very hard to make good predictions based onhuman behaviour.

Assume a given EV has an average stopover duration µ.Depending on the standard deviation σ of the predictionit can be determined how this influence the objective ofthe optimization model. Assuming that the data is normallydistributed, which may or may not be the case, a sample fromthe normal distribution N (µ, σ2) can be used to determinehow many times the linear model will produce false positiveestimates, i.e. predicting a EV is ready for charging whenit is not. This is done by only considering the prediction ofsome of the parameters of the optimization model, keepingin mind that true negatives are not considered bad estimateshere. Indeed these estimates are good by construction andcan be optimized further to yield less expensive results, i.e.minimizing the objective function further. However, the riskof getting false positives increases with the goal of minimizingthe objective function. To see how the errors may influencethe objective function, consider the surface w : U ⊂ R2 → R3

where the parameter space U is chosen to be the mean andstandard deviation of a series of predictions. The height usedhere is the probability of the EV being ready for charging aspredicted. This is expressed mathematically by

w(µ, σ) = P (a < x ≤ b), x ∼ N (µ, σ2).

In the current setting P (x ≤ b) = Fx(b) where

Fx(b) =1√2πσ2

∫ b

−∞e−

(µ−x)22σ2 dx

is the cumulative distribution function, it holds that

P (a < x ≤ b) = Fx(b)− Fx(a).

Fig. 4. Probability surface.

From this it is possible to deduce bounds for mean andstandard deviations from which the model works sufficientlywell. Let V ⊂ U be the set of parameters such that w(µ, σ) ≥0.95. Then the pairs (µ, σ) will belong to a closed and boundedregion V ⊂ R2.

Fig. 5. V cropped to the region [200; 1600]× [100; 300].

From this it can be concluded that predictions will get worseif:

1) For a fixed mean the standard deviation increases.2) For a fixed standard deviation the mean decreases.The estimator for the standard deviation of a sample x ∈ Rn

used here isσ =

1√n− 1.5

‖x− µ‖.

This estimator is not completely unbiased but it is sufficientfor the current analysis.

If the data for a particular EV is not normally distributedthen the above figures may look different. However it is likelythat the same considerations can be made based upon theparticular distribution.

VII. RESULTS

In the following the method described in the previoussection will be carried out on the AKTA data. The followingfigure illustrates that the predictions, in general, lie in thedesired region V .

Fig. 6. Results cropped to the region [200; 1600]× [100; 300].

Since the data set is quite small from a statistical viewpoint,it will be shown that these predictions may lie outside theregion V . Consider the 95% prediction interval

[µ− 1.96σ;µ+ 1.96σ].

From this the worst case scenario (assuming estimator for σ isgood) is that the predictions are placed in the lower part of theinterval. For the data set used here there are 205 observations

Fig. 7. Lower bound of the mean. Region [200; 1600]× [100; 300].

which have more than 10 days worth of observations. Thereare 13 observations which have a lower bound of the predictioninterval outside the region V . That is around 93.65% of thepredictions will lie in the region V . It is however unlikelythat all the predictions will be placed in the lower part of theprediction interval.

A example of how the optimization model performs whenthe variance of the predictions increases is now considered.The energy is assumed to be fixed, say E = 20000Wh, andthe prices αi have been sampled from a uniform distributionon the interval [80; 120].

α0 106.36 α10 99.58α1 88.41 α11 91.32α2 111.66 α12 112.49α3 114.88 α13 118.47α4 86.72 α14 110.87α5 105.41 α15 113.68α6 94.00 α16 105.70α7 105.81 α17 80.90α8 91.83 α18 97.57α9 110.45 α19 94.56

The possible charging periods βi are all assumed equal inlengths and measured in hours, i.e.

βi =1

2, ∀i.

The example demonstrates how the objective function isaffected when the stopover duration of initially 10 hours isincrementally cropped. This corresponds to increasing thestandard deviation of the prediction.

Fig. 8. Value of objective function.

In this example the crop is done by removing the first andlast elements of every parameter in the optimization model, seeequation 2. This can be different from actual events since thestandard deviation of last plugin prediction may be differentfrom that of first plugout prediction. The result shows theintuitive result that if the stopover duration decreases then thecost of charging is more.

A more thorough sensitivity analysis could be carried outon the parameters and these should at lease include chargingintervals, price time series and energy targets. Since datadescribing the energy targets was not available to the authorsat the time of writing this has not considered.

VIII. FURTHER EXTENSIONS OF THE MODEL

An interesting possibility is to let the EDISON EV aggre-gator target the ancillary services (A/S) markets. A/S marketsin general reward the capability of fast response which nicelymatches the capabilities of an aggregated group of EVs. InDenmark, markets such as ’Secondary reserve’ and ’Frequencycontrolled normal operation reserve’ are valid candidates for

EV participation, potentially offering a higher economic inven-tive than the earnings gained through smart charging alone. Afuture paper could seek to demonstrate how smart chargingand A/S can be combined by the aggregator to maximizethe overall economic value of controlled charging while stilladhering to all the constraints and considerations listed in thispaper.

IX. ADDITIONAL OPTIMIZATION CONSIDERATIONS

Although the optimization methods presented in this paperprimarily seeks to reduce the cost of energy, there are a greatmany additional parameters to be considered in the chargingof an EV. The primary sources of such considerations are thedistribution grid that connects the EV to the power system,the driving demands of the user of the EV and the internalbattery pack of the vehicle. The following will attempt to listthe most important of these considerations.

A. Distribution grid constraintsAnalyses of the impact of EVs on the power system have

revealed that without intervention, EV charging will placesubstantial pressure on the grid, particularly at the mediumand low voltage levels [1], [2], [3], [4]. The degree of gridcongestion is dependent on a number of factors includinglocal grid topology, penetration and distribution of EVs, andcharging management procedures [5]. Congestion will leadto reduced asset life expectancies, and in turn increasednetwork maintenance costs. Many studies have found thatcoordinated charging of vehicles is effective at reducing con-gestion, thereby increasing the penetration of EVs tolerableon any given network, and facilitating the deferral of gridreinforcement [1], [4].

Numerous approaches to dividing and managing availablegrid capacity are proposed within the literature. These can begenerally summarized as follows:

Static capacity allo-cation

Where each EV or aggrega-tor has a fixed portion of thecapacity allocated to them,i.e. a certain maximum al-lowed load in watts pr node.This approach is relativelyeasy to implement but willnot efficiently use the avail-able grid capacity.

Grid capacity market Where a dynamic market isimplemented in which theEVs, or the aggregators rep-resenting them, would bidfor certain amounts of theavailable capacity.

Dynamic tariff pricesignals

Aggregators or individualEVs respond to a price sig-nal that is representative ofthe congestion at each node,and optimize their chargingschedule accordingly.

Regardless of which of the above methods is used, an ag-gregator or fleet operator represents an efficient approach tocoordinate EV charging to avoid grid overload. In [10] a’Plug-In Hybrid Electric Vehicle (PHEV) manager’ is usedto coordinate all EV charging within a given zone, wherethe manager is aware of the grid capacity and can optimizecharging with respect to this. Alternatively, [9] proposes aniterative approach where by each aggregator coordinates thecharging of their fleet and reports this to the DSO.

A way to include grid considerations in the optimizationdefined in this paper is to limit the range of available chargingpowers at certain points in time.

If there is a need for incorporating grid constraints in model2 it may be useful to let p ∈ Rn×m

+ . Every row p is then thepossible charging configurations at time j where j = 1, . . . ,m.

B. Battery constraints

Another important area of research is on the lifetime ofEV batteries and the factors that influence it. Francesco et al.[18] has defined the following formula which summarizes themajor factors that has an influence on the battery lifetime L.

L = f(Cc, T,D,Cr),

where Cc is the charging cycles, T the temperature, D thedepth of discharge and Cr the charging rate. For most, if notall EVs, some sort of Battery Management System (BMS) willmonitor and to some degree control the above parameters as tomaximize the battery lifetime. One parameter relevant to theaggregator is the charging rate. For lithium Ion batteries, thecharging power should be reduced when the SOC exceeds acertain percentage. If this is not considered in the definition ofthe algorithm, less energy might be available at the beginningof a trip than defined by the energy target of the chargingschedule. One solution is simply to ignore the last roughlyten percent of the available energy capacity when generatinga charging schedule.

C. EV owner driving requirements

Driving predictions should be a tool to simplicity the EVowner’s management and control of smart charging. It should,however, always be possible for the EV user to assume directcontrol of the charging process. The main parameters that theEV user must be able to set are the energy demand and startingtime of the next desired trip. If such parameters are explicitlydefined by the user they will overwrite the values calculated bythe prediction algorithm. EV owner requirements can be meetby the optimization algorithm by simply setting the energytarget (E) and start time of the next trip to values defined bythe user.

At times, the EV owner might even wish to overwrite smartcharging all together, by choosing to instantaneously startcharging at the highest possible charging power. This woulddisable any kind of optimization.

X. CONCLUSION

This paper has presented the mathematical methods used bythe EV aggregator software developed in the EDISON project.

It has been demonstrated how a set of driving data fromcombustion engine vehicles in the Copenhagen area has beenused by a model to predict nightly stopovers. All of thepredicted stopover intervals lie in the 95% probability region(see figure 6). Considering the lower bound of the predictioninterval yields that 6.7% could lie outside the 95% probabilityregion (figure 7).

The prediction model is used by a suggested optimizationmodel as to minimize charging costs based on hourly spot mar-ket energy prices. It has been illustrated how the predictabilityof the data, measured on the standard deviation, will influencethe result of the objective function (figure 8).

Various additional parameters and constraints relating to thegrid, the EV owner and the EV battery have been discussedfor the sake of completeness, and it has been indicated howsome of these considerations can be added to the optimizationmodel.

Finally, ideas regarding further developments have beenmentioned. This includes adding revenue from acting on ancil-lary service markets to the savings earned by smart charging.

Figure 9 shows a screenshot from the aggregator softwareinterface. The screenshot includes three graphs that shows acharging schedule (bottom), the energy prices (middle) andthe predicted vs. measured availability (top) over a 24 hourperiod for a single simulated EV.

Fig. 9. Screenshot/EDISON EV aggregator.

REFERENCES

[1] A. Pecas Lopes, Soares, F.J., Rocha Almeida, P.M., Identifying Man-agement Procedures to Deal with Connection of Electric Vehicles in theGrid, presented at the Bucharest Power Tech Conference, IEEE 2009,Bucharest, Romania, 2009.

[2] K. Clement, Haesen, E., Driesen, J., Coordinated Charging of MultiplePlug In Hybrid Electric Vehicles in Residential Distribution Grids,presented at the Power Systems Conference and Exposition, 2009. PSCE’09. IEEE/PES Seattle, WA, USA, 2009.

[3] M. Arindam, Kyung, S.K., Taylor, J. , Giumento, A., Grid Impacts ofPlug-In Electric Vehicles on Hydro Quebec’s Distribution System, inTransmission and Distribution Conference and Exposition, 2010 IEEEPES New Orleans, USA, 2010.

[4] J. Taylor, Maitra, M., Alexander, D., Duvall, M, Evaluation of the Impactof Plug-In Electric Vehicle Loading on Distribution System Operations,presented at the Power & Energy Society General Meeting, 2009. PES’09. IEEE Calgary, AB, Canada, 2009.

[5] A. Maitra, Soo Kook, K., Taylor, J., Giumento, A., Grid Impacts of Plug-In Electric Vehicles on Hydro Quebec’s Distribution System, presented atthe Transmission and Distribution Conference and Exposition, 2010 IEEEPES New Orleans, LA, USA, 2010

[6] S. Babaeim, Steen, D., Tuan, L.A., Carlson, O., Bertling, L., Effects ofPlug-in Electric Vehicles on distribution systems: A real case of Gothen-burg, presented at the Innovative Smart Grid Technologies ConferenceEurope (ISGT Europe), 2010 IEEE PES Gothenburg, 2010.

[7] K. J. Dyke, Schofield, N., Barnes, M., The Impact of Transport Electrifi-cation on Electrical Networks, Industrial Electronics, IEEE Transactionson vol. 57, pp. 3917 - 3926, 08 February 2010 2010.

[8] Olle Sundstrom and Carl Binding, Planning Electric-Drive VehicleCharging under Constrained Grid Conditions. Accepted at InternationalConference on Power System Technology, China, October 24-28, 2010.Also available as IBM RZ 3785.

[9] J. A. Pecas Lopes, Soares, F.J., Rocha Almeida, P.M., Integration ofElectric Vehicles in the Electric Power System, Proceedings of the IEEEvol. 99, pp. 168- 183, 04 October 2010 2011.

[10] M. D. Galus, Andersson, G., Integration of Plug-In Hybrid ElectricVehicles into Energy Networks, presented at the 2009 IEEE BucharestPower Tech Conference, Bucharest, Romania, 2009.

[11] Carl Binding and Dieter Gantenbein and Bernhard Jansen and OlleSundstroem and Peter Bach Andersen and Francesco Marra and BjarnePoulsen and Chresten Traeholt, Electric Vehicle Fleet Integration inthe Danish EDISON Project - A Virtual Power Plant on the Island ofBornholm Proc. IEEE Power & Energy Society General Meeting 2010,Minneapolis, Minnesota, USA, July 25-29, 2010. Also available as IBMRZ 3761.

[12] Jacob Østergaard and Anders Foosnæs and Zhao Xu and Tim AndersMondorf and Claus Amtrup Andersen and Sven Holthusen and TorbenHolm and Maja Felicia Bendtsen and Kim Behnke, Electric Vehicles inPower Systems with 50% Wind Power Penetration : The Danish Caseand the EDISON programme. Presented at: European Conference Elec-tricity & Mobility. Wurzburg, Germany, 2009 In: European ConferenceElectricity & Mobility 2009

[13] Olle Sundstrom and Carl Binding, Optimization Methods to Plan theCharging of Electric Vehicle Fleets. Proc. ACEEE International Confer-ence on Control, Communication, and Power Engineering (CCPE), pp.323-328, Chennai, India, July 28, 2010. Also available as IBM RZ 3768.

[14] Olle Sundstrom and Carl Binding, Charging Service Elements for anElectric Vehicle Service Provider. IEEE Power & Energy Society GeneralMeeting, Accepted for publication, 2011.

[15] W. Kempton and J. Tomic, Vehicle-to-grid power implementation: Fromstatilizing the grid to supporting large-scale renewable energy. Journalof Power Sources, vol. 144, no. 1, pp. 280-294, June 2005.

[16] W. Kempton and J. Tomic, Vehicle-to-grid power fundamentals: Calcu-lating capacity and net revenue. Journal of Power Sources, vol. 144, no.1, pp. 268-279, June 2005.

[17] S. Kamboj and N. Pearre and W. Kempton and K. Decker and K.Trnka and C. Kern, Exploring the formation of Electric Vehicle Coalitionsfor Vehicle-To-Grid Power Regulation, in Proceedings of 2010 AAMASworkshop on Agent Technologies for Energy Systems.

[18] Marra, Francesco and Træholt, Chresten and Larsen, Esben and Wu,Qiuwei, Average Behavior of Battery - Electric Vehicles for DistributedEnergy System Studies, part of: IEEE Innovative Smart Grid TechnologyEurope 2010 (E-ISBN: 978-1-4244-8509-3) (DOI: 10.1109/ISGTEU-ROPE.2010.5638920) , pages: 1-7, 2010, IEEE Type: Full conferencepaper publ. in proceedings/book.

[19] Centre for Traffic and Transport (CTT). The AKTA Road PricingExperiment in Copenhagen, 2001-2003, unpublished raw data. TechnicalUniversity of Denmark.


F Facilitating a Generic Communication Interface to Dis-

tributed Energy Resources - Mapping IEC 61850 to

RESTful Services

This paper was published in the Proceedings of the First IEEE International Conference

on Smart Grid Communications (SmartGridComm), 2010, Gaithersburg, MD, USA.

160

Abstract—As the power system evolves into a smarter and

more flexible state, so must the communication technologies that

support it. A key requirement for facilitating the distributed

production of future grids is that communication and

information are standardized to ensure interoperability. The IEC

61850 standard, which was originally aimed at substation

automation, has been expanded to cover the monitoring and

control of Distributed Energy Resources (DERs). By having a

consistent and well-defined data model the standard enables a

DER aggregator, such as a Virtual Power Plant (VPP), in

communicating with a broad array of DERs. If the data model of

IEC 61850 is combined with a set of contemporary web protocols,

it can result in a major shift in how DERs can be accessed and

coordinated. This paper describes how IEC 61850 can benefit

from the REpresentational State Transfer (REST) service

concept and how a server using these technologies can be used to

interface with DERs as diverse as Electric Vehicles (EVs) and

micro Combined Heat and Power (µCHP) units.

I. INTRODUCTION

everal standards are currently under development that can

facilitate communication with DERs in a smart grid

constellation. Among these are OpenARD [1], OpenAMD

[2] and EMIX [3] as well as the IEC 61850 standard, on which

this paper will focus. The standard is thoroughly researched

and has been described in several papers [4] [5]. The standard

offers a structural decomposition of the units to which it

communicates. This means that each subcomponent of a DER

can be described by the information model of IEC 61850. This

makes the standard suitable for scenarios in which an

aggregator needs a fine-grained knowledge and control of the

state, structure and operation of a DER. The standard is a good

match for the virtual power plant concept, in which an

intermediate entity represents an aggregated group of DERs in

the power system and on the market. Apart from the syntactic

level, the standard also defines the protocols that carry the data

over a network. Without providing any recommendations on

the medium used, the standard proposes the use of the TCP/IP

protocols to enable internet communication. On the upper part

of the ISO OSI stack, the standard describes the use of the

Manufacturing Message Specification (MMS) standard as an

application-level protocol [6]. Replacing MMS with REST

services will, however, have certain advantages. A REST

service is a special flavor of web services which are connected

to the concept of a Service Oriented Architecture (SOA). As

REST services represent both a simple and well- documented

concept for achieving a high degree of interoperability, they

are a good candidate for use in the IEC 61850 protocol stack.

The next chapter describes IEC 61850 and REST in greater

detail. Then, the paper describes how an IEC 61850 server

was developed based on the above technologies, and finally a

case is defined for interfacing with a µCHP unit and an

electric vehicle. The IEC 61850 via REST implementation

suggested in this paper is used in the Danish EDISON project

[7], where a VPP-like aggregator should coordinate the

charging of electric vehicles, as well as in the Danish Generic

Virtual Power Plant project, where µCHPs are coordinated to

support the grid.

II. THE IEC 61850 STANDARD

The IEC-61850 standard was designed to enable

interoperability between different devices in the substation

environment, and to facilitate the adaptation of future

networking technologies. One of the cornerstones of the

standard is a layered data model (see Figure 1), which has

been designed to closely model the physical substation

environment.

A. Data model

The logical device is a virtual representation of a physical

device within the substation. It is comprised of a name, a path

to the object itself and a list of logical nodes. A logical device

contains one or more logical nodes, which represent various

components in the physical device.

As an example, the setup tested contains a logical node

called MMXN1. The MMXN part of the name refers to the type

of the node, which in this case is “Non-phase-related

measurements” [8]. Apart from the name, the MMXN class

must also contain all the data classes defined by the Logical

Node Class defined by the standard [8]. Beside this, instances

of the MMXN class can include a list of optional data classes

and attributes. The data classes represent meaningful

information inside the nodes and are declared recursively.

Facilitating a Generic Communication

Interface to Distributed Energy Resources Mapping IEC 61850 to RESTful Services

Anders Bro Pedersen Research Assistant

Technical University of Denmark

Einar Bragi Hauksson Research Assistant


Peter Bach Andersen PhD Student


Bjarne Poulsen Associate Professor


Chresten Træholt Associate Professor


Dieter Gantenbein Research Staff Member IBM Research, Zurich

S

Figure 1 – IEC 61850 data model hierarchy [9]

An attribute can either be a simple type, such as FLOAT32,

INT24 or BOOLEAN, or it can consist of both other simple

and complex attributes.

B. Data-Sets

The logical nodes can contain data-sets, which contain sets of

data that have a natural association [10]. Data-sets are

primarily used for reporting and logging [10], but can also be

requested directly. Included in the data-sets are properties

stating when a report/log should be triggered. An example of a

triggering condition is “Data Change”.

C. Logging and reporting

The IEC 61850 outlines a reporting mechanism, which is

essentially a payload-carrying event that is sent back to the

subscribing clients when triggered. Reporting is directly

linked to data-sets as it is the data attributes in the data-sets

that specify the trigger conditions. Closely related to reporting

is logging. They both rely on the data-sets for triggering, but

the reports are sent back to the clients, whereas the logs are

persisted locally [10].

D. Object references

Any object within the data model can be referred to directly

via its object path [10]. Because of the tree-like properties of

the model, this path resembles a fully-qualified file-name

notation. The path lists all the objects on the route from the

root of the model to the object in question. Where fully-

qualified file-name notation usually has a fixed delimiter

between object names, the IEC-61850 references use a slash to

separate the logical device from the rest of the path, which is

then separated by periods.

Figure 2 – Illustration of an IEC 61850 reference [9]

Included in the object reference is a filtering mechanism

called functional constraints (FC), which is used to group the

data-attributes. The functional constraints are usually specified

at the end of an object reference, incased in square brackets.

For the DC example, this would result in the path:

CHP1/MMXN1.Watt.mag.f[DC]

Because the paths resemble a fully-qualified file-name

notation, they can easily be mapped to a URL, which makes

the data model near perfect for REST. As the physical device

and the server are not referred in the object path, a REST URL

could look something like:

http://hostname/device/node/class/attribute

Included in IEC 61850 is also a service specification called

the Abstract Communication Service Interface (ACSI) [11],

which outlines a set of methods that are used when

communicating with the system. These methods have been

mapped to REST.

III. REST SERVICES

The REST architectural style was first described in 2000 in

a Ph.D. dissertation by Roy Fielding [12]. The industry did,

however, not embrace REST right away, probably because at

about the same time, the Single Object Access Protocol

(SOAP) [13] was embraced by most of the large software

vendors and therefore got a lot of attention. As time passed

and SOAP grew, adding numerous extensions, people started

looking for a lighter, more web-centric alternative. Today

many big web companies provide REST services for others to

interact with their systems. These include companies such as

Amazon, Google, Yahoo and Facebook.

Where SOAP comes complete with a seemingly ever-

growing suite of extensions, the primary aim of REST is to

stick closer to the basic functionality of the HTTP protocol on

which the web is based. Unlike SOAP, there is no standard

available that describes REST. Instead, developers should

follow the REST principles when creating a RESTful [14]

service.

The most important REST principle is to expose the

resources in a RESTful service as unique URLs. An example

of this might be a library service where the whole book

catalog could be accessed by using the URL

http://library.com/books/ whereas a single book could be

accessed using the URL http://library.com/books/123 (where

Substation(physical device)

Server

Logical Device

Logical Node

Data Object

Common Data Classes

Common Attributes

Standard Data Types

1

n

1

n

1

n

1

n

1

n

1

n

1

n

n..1

1..n

Lo

gic

al m

od

el

CHP1 MMXN1 Watt mag f

Logical device

Logical node

Data class

Data attribute

Data attribute

123 is the ID of the book). As opposed to SOAP services,

where you would issue a method call such as createBook,

deleteBook, RESTful services utilize the HTTP methods,

GET, POST, PUT and DELETE for reading, creating,

updating and deleting resources, respectively.

The REST principles do not define any specific format for

request or response data. Most common, however, is the XML

format, but other formats such as JavaScript Object Notation

(JSON) [15] are becoming increasingly popular, especially in

Asynchronous JavaScript And XML (AJAX) services. Ideally

the client is able to ask for specific representation by setting

the Accept request header to the desired format name.

Like the HTTP protocol itself, RESTful services are

stateless, meaning that no state should be stored on the server

between requests from the client. Each request should

therefore contain all the information necessary to serve the

client. RESTful services also embrace other aspects of the

HTTP protocol, such as status codes, conditional get and

caching.

Status codes are sent along with any HTTP response and

indicate the type of the response. Examples of common status

codes are 200 (OK) for a successful request and 404 (NOT

FOUND) for a request for a non-existing resource. HTTP

defines status codes in the range 1xx to 5xx.

Conditional Get allows the client to ask the server whether

the requested resource has changed since it was last retrieved.

This is achieved either by sending the If-Modified-Since

header with its value set to the time of the last retrieval or by

sending the eTag header that came with the last response. If

the resource has not been modified the server only returns

status code 304 (NOT MODIFIED). This can be important for

performance, as unchanged data does not need to be re-sent.

Another closely related header is the cache header, sent

along with the response from the server. The cache header

allows the server to tell the client whether and for how long

the client should cache the response. This improves

performance as it saves the client from requesting the same

data again, if it is not expected to change. This is mainly used

for static resources, such as images on the web, but can also be

utilized in RESTful services.

The functionality of RESTful services is very closely

related to the functionality of the web, which has been

extremely successful over the last couple of decades. This is

mainly due to its scalability, interoperability and the fact that it

is simple and easy to understand. HTTP and the WWW are

used for a wide variety of tasks ranging from personal home

pages to secure internet-banking and e-commerce. HTTP

already has a built-in authentication mechanism but since

many, if not all of the tasks mentioned call for strict security,

several protocols have developed to add things like encryption

to HTTP. Among some of the more well-known ones are open

standards, such as the TSL/SSL [16] protocols. Because of

technologies like these and the needs from which they arose,

the idea of using REST for communication in electrical

systems, as is the case with the VPP and the distributed energy

resources, could surely be considered safe.

IV. RESTFUL INTERFACE FOR IEC 61850

As mentioned, the IEC 61850 reference paths resemble a

fully-qualified file-name notation or a URL. This simplifies

the task of creating a RESTful interface for the IEC 61850

data model. The idea is that the various objects in the data

model hierarchy can be thought of as resources, which can be

accessed by using the IEC 61850 object reference. As an

example the data attribute

CHP1/MMXN1.Watt.mag.f

belonging to logical device “CHP1”, logical node “MMXN1”

and so on would have the URL

CHP1/MMXN1/Watt/mag/f

and the entire “Watt” data object could be retrieved with the

URL CHP1/MMXN1/Watt. For getting data, as in the example

above, an HTTP GET request would be used. The server

would then respond with the requested data in XML format as

shown in Figure 3. For setting data, the same URL would be

used, but with the HTTP method POST instead of GET.

Figure 3 - REST communication overview.

The ACSI interface described in IEC 61850-7-2 enables

clients to inspect the data model, to read and write data, and to

access data-sets, logs and more. To do so, the client calls

methods such as GetDataValues and SetDataValues. These

methods resemble traditional methods in a programming

language with input arguments and return values. As described

above, the REST architecture guidelines define services as a

set of resources instead of methods as in the case of ACSI.

To make a resource-oriented interface for the IEC 61850

standard, a mapping from the ACSI methods to URL and

HTTP method-pairs has been defined. The mapping for the

data model of the IEC 61850 standard is shown in Table 1.

Mappings for data-sets, reporting, logging, setting-groups

and substitutions can be made in a similar manner. These

mappings are presented in full detail in [9].

The first column in Table 1 shows the URL template of a

resource, the second column shows the HTTP method used to

get or set the URL, and the third column shows the ACSI

services this resource replaces. The response from the GET

requests to the resources shown in Listing 1 can be modified

by using query string parameters. Three query string

parameters have been defined: expandLevel, fc and

includeValues.

The expandLevel parameter controls how big a portion of

the data model is retrieved. The default value of this parameter

is zero, which results in only the requested level of the data

ServerRequest

Clients

RE

ST

Inte

rfa

ce

VPP

Monitor

...

Response

CHP1/MMXN1/Watt/mag/f

XML

IEC

61

85

0

Se

rve

r

model being retrieved, as can be seen in Listing 1. Setting the

expandLevel parameter to a higher number or to the value

“all”, allows the client to retrieve bigger portions of the data-

model hierarchy, as seen in Listing 2. By retrieving the whole

data model, the client can easily inspect an entire logical

device and a single URL can therefore replace multiple ACSI

services.

The two remaining parameters accepted by the RESTful

service are fc and includeValues. The fc parameter is used for

filtering the data-model hierarchy to include only data

attributes with the specified functional constraint. The

includeValues parameter specifies whether the data values

should be included in the response in addition to the data

model.

TABLE 1 - ACSI TO REST MAPPING

Url Meth. ACSI equivalent

/ GET GetServerDirectory (GetAllDataValues)

(GetDataValues)

/[LD]/ GET GetLDDirectory

(GetAllDataValues)

(GetDataValues)

/[LD]/[LN] GET GetLNDirectory (GetAllDataValues)

(GetDataValues)

/[LD]/[LN]/[DO] GET GetDataDirectory

GetDataDefinition (GetAllDataValues)

(GetDataValues)

/[LD]/[LN]/[DO]/[DA] GET (GetAllDataValues)

(GetDataValues)

/[LD]/[LN]/[DO]/[DA] POST SetDataValues

As Figure 3 illustrates, the response format of the RESTful

service is in XML. Because of XMLs ability to model

arbitrary data structures it is possible to represent the

hierarchical IEC 61850 data model in a concise and readable

format.

To show this, two URLs and their resulting output are

given. Listing 1 shows a response containing a single value;

Listing 2 is an example of what a larger view of the data

model might look like.

1. <DA Name="f" Type="FLOAT32"

Ref="EV1/ZCEV1/MaxRtDchPwr/setMag/f">16800</DA>

Listing 1 - Response for URI: /EV1/ZCEV1/MaxRtDchPwr/setMag/f

As seen in Listing 2, a request for an entire data hierarchy

can be quite verbose and might not be suitable for frequent use

in a production environment. This kind of request is, however,

very useful during the configuration and development phase

when new devices need to be discovered. As the RESTful

interface uses standard HTTP GET request, it is even possible

to use a web browser to discover an IEC-61850-enabled

device. As with ACSI methods such as GetServerDirectory

and GetLDDirectory, the RESTful interface also enables

clients to programmically discover devices in a generic

manner.

1. <DO Name="MaxRtDchPwr" Type="ASG"

Ref="EV1/ZCEV1/MaxRtDchPwr">

2. <DA FC="SP" Name="setMag" Type="Struct"

Ref="EV1/ZCEV1/MaxRtDchPwr/setMag">

3. <DA Name="f" Type="FLOAT32"

Ref="EV1/ZCEV1/MaxRtDchPwr/setMag/f">16800</DA>

4. </DA>

5. <DA FC="CF" Name="units" Type="Struct"

Ref="EV1/ZCEV1/MaxRtDchPwr/units">

6. <DA Name="SIUnit" Type="Enum"

Ref="EV1/ZCEV1/MaxRtDchPwr/units/SIUnit">Watt</DA>

7. <DA Name="multiplier" Type="Enum"

Ref="EV1/ZCEV1.MaxRtDchPwr/units/multiplier">Item</DA>

8. </DA>

9. <DA FC="DC" Name="dU" Type="Unicode255"

Ref="EV1/ZCEV1/MaxRtDchPwr/dU">Maximum rated discharging

power</DA>

10. </DO>

Listing 2 - Response for URI: /EV1/ZCEV1/MaxRtDchPwr/?expandLevel=all

Although XML has been chosen as the output format in the

example above, the RESTful interface could just as well use

other output formats. Thanks to the Accept header discussed

above, the same RESTful service could both accept and output

XML and JSON. One argument for using the JSON format is

that it is designed for serializing common objects, lists and

scalar values found in virtually all programming languages

and can therefore easily be deserialized by those languages.

And without the verbose declarations, it is definitely more

compact. JSON libraries exist for all major programming

languages/frameworks, including Java, .Net, C++, Python etc

[17].

As discussed in Section II, the IEC 61850 standard defines a

reporting mechanism. Because of the connection-less nature of

HTTP, and thereby of REST, implementing reporting

callbacks to the client is not entirely straightforward.

However, such mechanisms exist, for example the WebHooks

model [18] proposed by Jeff Lindsay, NASA Ames Research

Center. The approach is to have the IEC 61850 server enable

clients to subscribe to reporting events, similar to the approach

described in IEC 61400-25-3[19], and pass along with that

subscription request a callback URL which the IEC server can

use for sending reports to the client as HTTP POST requests.

Security must be kept in mind using this approach, because the

client must be able to validate that the callbacks received do

indeed come from the server. This could be solved using

HTTP-basic authentication or X.509 client certificates [20].

V. CASE STUDIES

To test the usefulness of REST for mapping to the IEC

61850 standard, a server complying with both of these has

been implemented. The server was designed “from scratch” to

support multiple devices and to test the scalability of the

REST implementation. The server has been written entirely in

C# for the .NET framework.

To facilitate the deployment to embedded devices and add

support for non-Windows systems, a small open-source in-

process web server was chosen to host the REST interface.

Except for the logging and reporting mechanisms, which run

in the background, the server waits for requests from potential

clients. When a request is received for a given URL, the server

looks up the requested object in its internal representation of

the data model. If the request is a “write” the server will

simply return a status code as confirmation. In the case of a

“read”, however, the requested object is serialized to XML

and returned to the client. One of the benefits of this approach

is that it is relatively simple to return not just the requested

object but the entire sub-model. In fact, if a request is received

for the logical device the server is able to return the entire data

model, including all the values, using a single response.

The goal of these case studies is to test the REST mapping

against different DERs, namely,

to describe the µCHP setup and briefly show how the

communication has been mapped

to touch upon the challenges of mapping the electric

vehicles and what was needed to accomplish this.

A. Case study: Interfacing with a µCHP

The server was designed using a modular plug-in architecture

with a generic interface, which enables easier adaptation and

installation of virtually any type of device.

Among the IEC-61850-enabled devices is a pair of Dachs

µCHP units from Senertec. The Dachs accepts a series of

different commands over the RS232 line and each of these

commands returns a different set of values. These values

include electrical and temperature measurements. When a

complex REST request is received, i.e., more than one data

attribute is requested, the IEC 61850 server requests each

required value from the Dachs module. The Dachs module

then locates the requested value in the set of values returned

by each of its commands. As the values are requested from the

Dachs module one at a time, the module caches the result of

each command for a fraction of a second to avoid having to

issue a command multiple times for each REST request.

Figure 4 - µCHP module RS232 communication.

Two client applications were developed to demonstrate the

system and to facilitate testing. One of these demo

applications is a web client showing the various devices

attached to the server, as well as all the measurements and

controls available. A screenshot from the client showing one

of the µCHPs can be seen in Figure 5.

Figure 5 - Screenshot of the web client.

B. Case study: Interfacing with an EV

Another example of a plug-in made for the server is for an

EV charging-spot. As no real vehicles were available at the

time of development, the plug-in was mapped to an EV

simulation instead, which further illustrates the versatility of

the system. The simulation includes both a charging-spot and

an EV, which contains a simulated battery. When the EV is

plugged into the charging spot an EV aggregator-server can

tell the charging spot to control the charging using schedules

sent via the IEC 61850 REST interface. Such an EV

aggregator-server is used in the EDISON project [21].

Figure 6 - Setup showing simulated EV and charging-spot.

The charging spot therefore acts as a proxy for the EV. This

is in accordance to [9] and [21] and means that the EV does

not need a wireless connection for being utilized by the

aggregator server. Because of this, even though the charging

spot and the vehicle are simulated as individual entities, they

are mapped with the charging spot as the logical device and

the currently connected EV represented as a logical node. The

aggregator server, or any other client, can read various values

from the charging-spot and the EV via the IEC 61850

interface, such as the current power usage, the phases in use

and the current state of charge.

Command

Result Set

Server Device

µCHP

Da

ch

s P

lug

-in

RS232

RS232

IEC

61

85

0

Se

rve

r

Server

Sim

ula

tio

n

Plu

g-in

IEC

61

85

0S

erv

er

Simulated Charging-spot and EV

There is an addition to the IEC 61850 standard, the IEC

61850-7-420 [22], that deals specifically with DERs. After

having analyzed the requirements for the charging spot and

EV, it became apparent that this standard needed further

extension. Logical nodes for charging-spots (ZCHS) and for

EVs (ZCEV) have been defined. These nodes contain essential

attributes such as the state-of-charge of the EV, power limits

and battery capacity. For controlling charging and discharging

of the EV, the logical nodes for energy schedule, defined in

[22], have been utilized. Since these extensions are outside the

scope of this paper and subject for a later publication they will

not be further described here.

Figure 7 - Demo showing the current state of an EV.

Together with the EV simulation, a small graphical client

was developed to show the current state of the EV and the

charging spot. This application, which is shown in Figure 7,

displays a small car that moves between the road and the

buildings, representing one of the three possible states “in

use”, “idle” and “plugged-in”.

C. Case study conclusion

Although µCHPs and EVs differ significantly in both function

and composition, the cases presented in this section show how

IEC 61850 with the REST interface can support

communication with both types of DERs. The units can be

monitored and controlled to optimize their behavior in relation

to energy prices, user requirements and the state of the power

system.

VI. CONCLUSION

The main aim of this paper has been to demonstrate how

RESTful services, in conjunction with the data model of IEC

61850, can be used to increase interoperability and simplicity

in DER communication. The IEC 61850 standard has a large

and well-defined set of logical notes describing the various

components and values of a DER unit. This paper has shown

that the object reference path of the IEC 61850 data model can

easily be mapped to the URL format in the resource-oriented

approach used by REST. Besides offering an intuitive way of

accessing information, REST also provides better

interoperability and simplicity by shedding much of the

complexity present in SOAP-based web services. The

advantages of using IEC 61850 with REST have been

demonstrated by building an IEC 61850 server and describing

its functionality in two case studies.

There is no doubt that the ambition of actively integrating

DERs into a smart grid constellation will rely on the

utilization of contemporary web concepts and standards. This

paper serves as an input to the identification of the ICTs

capable of satisfying the communication requirements for the

power system of the future.

REFERENCES

[1] OPENADR - Outreach Collaborative homepage.

http://www.openadrcollaborative.org/

[2] OPENAMD – standard sponsor. UCA International Users Group.

http://www.ucaiug.org/

[3] EMIX – standard sponsor. OASIS Blue Initiative. http://www.oasis-blue.org/

[4] "Communication networks and systems in substations," Commission

Electrotechnique Internationale (IEC), vol. 1, July 2003.

[5] S. Heights, R. Mackiewicz, "Technical overview and benefits of the IEC

61850," Oct. 2006.

[6] "Manufacturing Message Specification (MMS)", Partl:Service Definition, Part2: Protocol Specification,International Standard IS0 9506, 1989.

[7] Edison project homepage. http://www.edison-net.dk/

[8] "IEC 61850-7-4, Basic communication structure - Compatible logical node

classes and data object classes".

http://webstore.iec.ch/Webstore/webstore.nsf/ArtNum_PK/38211

[9] E. B. Hauksson, A. B. Pedersen, "Enabling distributed energy resources in

a virtual power plant using IEC-61850," Department of Informatics and

Mathematical Modeling at the Technical University of Denmark, Master Thesis 2010.

[10] "IEC 61850-7-1, Basic communication structure for substation and feeder

equipment Principles and models". http://webstore.iec.ch/Webstore/webstore.nsf/ArtNum_PK/30928

[11] "IEC 61850-7-2, Basic communication structure for substation and feeder

equipment - Abstract communication service interface (ACSI),". http://webstore.iec.ch/webstore/webstore.nsf/Artnum_PK/30581

[12] R. T. Fielding, "Architectural Styles and the Design of Network-based

Software Architectures," PhD Thesis 2000. University of California, Irvine

[13] SOAP Specification. http://www.w3.org/TR/soap/

[14] IBM developerWorks, A. Rodriguez. RESTful Web services: The basics.

https://www.ibm.com/developerworks/webservices/library/ws-restful/

[15] The application/json Media Type for JavaScript Object Notation.

http://tools.ietf.org/html/rfc4627

[16] The Transport Layer Security (TLS) Protocol.

http://tools.ietf.org/html/rfc5246

[17] JSON. http://www.json.org/

[18] WebHooks. http://www.webhooks.org/

[19] "IEC 61400-25-3, Communications for monitoring and control of wind

power plants - Information exchange models".

[20] Internet X.509 Public Key Infrastructure: Certification Path Building. http://tools.ietf.org/html/rfc4158

[21] Carl Binding, Dieter Gantenbein, Bernhard Jansen, Olle Sundström,

Peter Bach Andersen, Francesco Marra, Bjarne Poulsen, and Chresten Træholt, "Electric Vehicle Fleet Integration in the Danish EDISON Project

- A Virtual Power Plant on the Island of Bornholm," IEEE Power &

Energy Society, 2010 General Meeting , July 25-29, 2010, Minneapolis, Minnesota, USA.

[22] "IEC 61850-7-420, Basic communication structure - Distributed energy

resources logical nodes". http://webstore.iec.ch/Webstore/webstore.nsf/Artnum_PK/42706

References

[1] IEEE smart grid website

. http://smartgrid.ieee.org/ieee-smart-grid. 2

[2] C. Kieny, B. Berseneff, N. Hadjsaid, Y. Besanger, and

J. Maire. On the concept and the interest of vir-

tual power plant: Some results from the European

project Fenix. In Proc. IEEE Power and Energy Society

General Meeting, 2009. 3, 59

[3] You Shi, Chresten Traeholt, and Bjarne Poulsen. A

market-based virtual power plant. In Proc. Inter-

national Conference on Clean Electrical Power (ICCEP),

2009. 3

[4] You Shi, Chresten Traeholt, and Bjarne Poulsen.

Generic virtual power plants: Management of dis-

tributed energy resources under liberalized elec-

tricity market. In Proc. 8th IET International Confer-

ence on Advances in Power System Control, Operation and

Management., 2009. 3, 59, 60

[5] Carl Binding, Dieter Gantenbein, Bernhard Jansen, Olle

Sundstroem, Peter Bach Andersen, Francesco Marra,

Bjarne Poulsen, and Chresten Traeholt. Electric ve-

hicles as a new power source for electric utilities.

In Proc. IEEE Power and Energy Society General Meeting

2010, 2010. 3

[6] Willett Kempton and Steven Letendre. Electric vehi-

cles as a new power source for electric utilities.

Trans. Res. D 2 (1997), 2(3):pp. 157–175, 1997. 4

[7] Burden of disease from environmental noise -

Quantification of healthy life years lost in Europe.

Technical report, World Health Organisation (WHO),

2011. Download from http://www.who.int. 6

[8] Luftforurening med partikler i Kbenhavn. Techni-

cal report, Danmarks Miljundersgelser (DMU), Aarhus

university, 2003. Download from http://dce.au.dk. 6

[9] Dansk elbil komite website

. http://www.danskelbilkomite.dk. 6

[10] Effektiv anvendelse af vindkraftbaseret el i Dan-

mark. Technical report, Energinet.dk, 2009. Download

from http://www.energinet.dk. 7

[11] Smart Grid i Danmark. Technical report, En-

erginet.dk and Danish Energi Association. Download

from http://www.energinet.dk. 7

[12] EDISON website

. http://www.edison-net.dk. 8

[13] E-Mobility Berlin project, RWE project

description

. https://www.rwe-mobility.com/web/cms/de/1158022/emobility/.

8

[14] VATTENFALL project description (in german)

. http://www.vattenfall.de/de/wirtschaftsverkehr.htm.

8

[15] MERGE project website

. http://www.ev-merge.eu/. 8

[16] University Of Delaware, V2G website

. http://www.udel.edu/V2G/. 8

[17] IEC website

. http://www.iec.ch/. 9

[18] CENELEC website

. http://www.cenelec.eu/. 9

[19] EcoGrid EU website

. http://www.eu-ecogrid.net/. 20

[20] Jasna Tomi and Willett Kempton. Using fleets of

electric-drive vehicles for grid support. Journal of

Power Sources, 168:459468, 2007. 21, 34

[21] Willett Kempton, Victor Udo, Ken Huber, Keven

Komara, Steve Letendre, Scott Baker, Doug Brun-

ner, and Nathaniel Pearre. A Test of Vehicle-

to-Grid (V2G) for Energy Storage and

Frequency Regulation in the PJM Sys-

tem. Test report, University Of Delaware, 2008.

http://www.udel.edu/V2G/ArticlesandPapers.html.

21, 25

[22] N. Rotering and M. Ilic. Optimal charge control

of plug-in hybrid electric vehicles in deregulated

electricity markets. IEEE Transactions on Power Sys-

tems, 2010. 21

[23] T. Kristoffersen, K. Capion, and P. Meibom. Optimal

charging of electric drive vehicles in a market en-

vironment. Applied Energy, page 19401948, 2010. 21

[24] Bjarke Heesche. Algoritmer til prisstyret opladning

af elbiler. Bachelor report, Technical University Of

Denmark, Department of Electrical Engineering, 2012.

http://www.dtu.dk/centre/cet/English/education/. 21

[25] Dieter Gantenbein. WP3 Aggregation - The Advan-

tage of Coordinated Consumption Planning. EDI-

SON project website: http://www.edison-net.dk/. Tech-

nical web article. 32

[26] Qiuwei Wu, Jakob Munch Jensen, Lars Henrik Hansen, An-

dreas Bjerre, Arne Hejde Nielsen, and Jacob stergaard.

EV portfolio management and grid impact study.

Project report, EDISON, 2012. http://www.edison-

net.dk/Dissemination/Reports/. 33

[27] Andr G. Madureira, Andreas F. Raab, Andrs Ramos, Evan-

gelos Karfopoulos, Fernando Bez Chicharro, Jess La-

torre, Jianzhong Wu, Kostas Tsatsakis, Kristin Dietrich,

Manolis Voumvoulakis, Marco A. M. Saran, Mauro Rosa,

Ricardo Ferreira, Jos Lus Meirinhos, Ivo Costa, Mixalis

Stamoulis, Mohsen Ferdowsi, and Yunfei Mu. Evalua-

tion of the impact that a progressive deployment

167

REFERENCES

of EV will provoke on electricity demand, steady

state operation, market issues, generation sched-

ules and on the volume of carbon emissions. Wp

3 task 3.2 part ii of deliverable d3.2, MERGE, 2012.

http://www.ev-merge.eu/. 33

[28] Nan Shao. The Impact of Integration of Distributed

Energy Resources in a Danish Low-Voltage Dis-

tribution Network. Master thesis, Technical Univer-

sity Of Denmark, Department of Electrical Engineering,

2012. 33

[29] Francesco Marra, Morten Mller Jensen, Rodrigo Garcia-

Valle, Chresten Trholt, and Esben Larsen. Power Qual-

ity Issues into a Danish Low-Voltage Grid with

Electric Vehicles. In Proc. 11th International Confer-

ence on Electrical Power Quality and Utilisation (ISBN:

978-1-4673-0379-8), 2012. 33

[30] Shengnan Shao, Manisa Pipattanasomporn, and Saifur

Rahman. Demand Response as a Load Shaping Tool

in an Intelligent Grid With Electric Vehicle. 2011

IEEE Transactions on Smart Grid, 2, Issue 4:624–631,

2011. 34

[31] Niamh O’Connell, Qiuwei Wu, Jacob stergaard,

Arne Hejde Nielsen, and Seung-Tae Cha. Electric

Vehicle (EV) Charging Management with Dy-

namic Distribution System Tariff. In Proc. 2011

IEEE PES Innovative Smart Grid Technologies, 2011.

34, 36

[32] Olle Sundstroem and Carl Binding. Flexible Charg-

ing Optimization for Electric Vehicles Consider-

ing Distribution Grid Constraints. In Proc. 2012

IEEE PES, Power Engineering Society General Meeting,

2012. 34, 82

[33] Thomas Christian Jensen, Ole Kveiborg, and An-

ders Fjendbo Jensen. Vil bilisterne koebe elbiler?

Trafik og Veje (ISSN: 1903-7384), 12:4243, 2011. 42

[34] Nathaniel S. Pearre, Willett Kempton, Randall L.

Guensler, and Vetri V. Elango. Electric vehicles:

How much range is required for a days driv-

ing? Transportation Research 2011, ELSEVIER, page

11711184, 2011. available at ScienceDirect. 43

[35] Qiuwei Wu, Arne Hejde Nielsen, Jacob stergaard, Seung-

Tae Cha, Francesco Marra, Yu Chen, and Chresten

Trholt. Driving Pattern Analysis for Electric Ve-

hicle (EV) Grid Integration Study. In Proc. 2010

IEEE Inovative Smart Grid Technology Europe, 2010. 44

[36] Olle Sundstroem, Olivier Corradi, and Carl Binding. To-

ward Electric Vehicle Trip Prediction for a Charg-

ing Service Provider. In Proc. 2012 Electric Vehicle

Conference (IEVC), IEEE International, 2012. 44

[37] R. Simmons, B. Browning, Y. Zhang, and V. Sadekar.

Learning to predict driver route and destination

intent. In Proc. 2006 IEEE Intelligent Transportation

Systems Conference, 2006. 44

[38] Peter Cocrona, Franziska Bhlera, Isabel Neumanna,

Thomas Frankea, Josef F. Kremsa, Maximilian Schwalmb,

and Andreas Keinathb. Methods of evaluating elec-

tric vehicles from a users perspective the MINI E

field trial in Berlin. IET Intelligent Transport Systems

(ISSN: 1903-7384), 5(2):127–133, 2011. 44

[39] Sachin Kamboj, Willett Kempton, and Keith S. Decker.

Deploying Power Grid-Integrated Electric Vehi-

cles as a Multi-Agent System. In Proc. Tenth In-

ternational Joint Conference on Autonomous Agents and

Multiagent Systems, 2011. 59, 61

[40] S.A. Klein. Security, cost, and operational bene-

fits of IEC-61850. In proc. 2008 IEEE Power and En-

ergy Society General Meeting - Conversion and Delivery

of Electrical Energy in the 21st Century. 68

[41] R.E. Mackiewicz. Overview of IEC 61850 and Bene-

fits. In Proc. IEEE PES 2006 Power Systems Conference

and Exposition, pages 623–630, 2006. 68

[42] T.S. Ustun, C. Ozansoy, and A. Zayegh. Modeling of

a Centralized Microgrid Protection System and

Distributed Energy Resources According to IEC

61850-7-420. IEEE 2012 Transactions on Power Sys-

tems, 27(3):1560–1567, 2012. 68

[43] N. Honeth, Wu Yiming, N. Etherden, and L. Nordstrom.

plication of the IEC 61850-7-420 data model on a

Hybrid Renewable Energy System. In Proc. IEEE

2011 Trondheim PowerTech. 68

[44] Jens Schmutzler, Sven Groening, and Christian Wietfeld.

Management of Distributed Energy Resources in

IEC 61850 using Web Services on Devices. In Proc.

2011 IEEE SmartGridComm, 2011. 68

[45] Stjepan Sui and Ante Martini. Utilizing SOA-ready

Devices for Virtual Power Plant Control in

Semantic-enabled Smart Grid. In Proc. 2011 IEEE

SmartGridComm, 2011. 69

[46] Joachim Skov Johansen. Electric Vehicle Communica-

tion Methods. Special project report, Technical Uni-

versity Of Denmark, Department of Electrical Engineer-

ing, 2012. Report available through CEE, DTU (contact

thesis author). 70

[47] E. B. Hauksso and A. B. Pedersen. Enabling distributed

energy resources in a virtual power plant using

IEC-61850. Master thesis, Department of Informatics

and Mathematical Modeling at the Technical University

of Denmark, 2010. 71, 72, 73

[48] Roy T. Fielding. Architectural Styles and the Design

of Network-based Software Architectures. Phd the-

sis, University of California, Irvine, 2000. 71

[49] Bernhard Jansen, Carl Binding, Olle Sundstroem, and Di-

eter Gantenbein. Architecture and Communication

of an Electric Vehicle Virtual Power Plant. In Proc.

2010 IEEE International Conference on Smart Grid Com-

munications,, 2010. 76

[50] Olle Sundstroem and Carl Binding. Optimization

methods to plan the charging of electric vehicle

fleets. In Proc. 2012 International Conference on Con-

trol, Communication and Power Engineering, 2010. 81

[51] Olle Sundstroem, Olivier Corradi, and Carl Binding.

Charging Service Elements for an Electric Vehi-

cle Service Provider. In Proc. 2011 IEEE Power and

Energy Society General Meeting,, 2011. 81

168

REFERENCES

[52] Olle Sundstroem and Carl Binding. Planning Electric-

Drive Vehicle Charging Under Constrained Grid

Conditions. In Proc. 2010 International Conference on

Power System Technology, 2010. 81

[53] J. Lopes, F. Soares, and P. Almeida. Identifying man-

agement procedures to deal with connection of

electric vehicles in the grid. In Proc. 2009 IEEE Pow-

erTech, 2010. 81

169

REFERENCES

www.cee.dtu.dkDepartment of Electrical EngineeringCentre for Electric Power and Energy (CEE)Technical University of DenmarkElektrovej 325DK-2800 Kgs. LyngbyDenmark

Tel: (+45) 45 25 35 00Fax: (+45) 45 88 61 11E-mail: [email protected]

170

Documents

Intelligent Electric Vehicle Integration - Domain Interfaces and